You are on page 1of 32

Environment International 169 (2022) 107468

Contents lists available at ScienceDirect

Environment International
journal homepage: www.elsevier.com/locate/envint

Full length article

Systematic evidence map (SEM) template: Report format and methods used
for the US EPA Integrated Risk Information System (IRIS) program,
Provisional Peer Reviewed Toxicity Value (PPRTV) program, and other “fit
for purpose” literature-based human health analyses
Kristina A. Thayer a, *, Michelle Angrish a, Xabier Arzuaga a, Laura M. Carlson b, Allen Davis a,
Laura Dishaw a, Ingrid Druwe a, Catherine Gibbons a, Barbara Glenn a, Ryan Jones b,
J. Phillip Kaiser a, Channa Keshava a, Nagalakshmi Keshava a, Andrew Kraft a, Lucina Lizarraga a,
Amanda Persad a, Elizabeth G. Radke a, Glenn Rice a, Brittany Schulz c, Rachel M. Shaffer a,
Teresa Shannon a, Andrew Shapiro b, Shane Thacker b, Suryanarayana V. Vulimiri a,
Antony J. Williams d, George Woodall b, Erin Yost a, Robyn Blain e, Katherine Duke e,
Alexandra E. Goldstone e, Pam Hartman e, Kevin Hobbie e, Brandall Ingle e, Courtney Lemeris e,
Cynthia Lin e, Alex Lindahl e, Kristen McKinley e, Parnian Soleymani e, Nicole Vetter e
a
Center for Public Health and Environmental Assessment, Chemical & Pollutant Assessment Division, US EPA, DC, USA
b
Center for Public Health and Environmental Assessment, Health & Environmental Effects Assessment Division, US EPA, NC, USA
c
Oak Ridge Associated Universities, TN, USA
d
Center for Computational Toxicology and Exposure, US EPA, NC, USA
e
ICF, VA, USA

A R T I C L E I N F O A B S T R A C T

Handling Editor: Adrian Covaci Background: Systematic evidence maps (SEMs) are gaining visibility in environmental health for their utility to
serve as problem formulation tools and assist in decision-making, especially for priority setting. SEMs are now
Keywords: routinely prepared as part of the assessment development process for the US Environmental Protection Agency
Systematic review (EPA) Integrated Risk Information System (IRIS) and Provisional Peer Reviewed Toxicity Value (PPRTV) as­
Risk assessment
sessments. SEMs can also be prepared to explore the available literature for an individual chemical or groups of
Hazard characterization
chemicals of emerging interest.
Toxicity
Scoping review Objectives: This document describes the typical methods used to produce SEMs for the IRIS and PPRTV Programs,
Systematic evidence map as well as “fit for purpose” applications using a variety of examples drawn from existing analyses. It is intended to
serve as an example base template that can be adapted as needed for the specific SEM. The presented methods
include workflows intended to facilitate rapid production. The Populations, Exposures, Comparators and Out­
comes (PECO) criteria are typically kept broad to identify mammalian animal bioassay and epidemiological
studies that could be informative for human hazard identification. In addition, a variety of supplemental content
is tracked, e.g., studies presenting information on in vitro model systems, non-mammalian model systems,

* Corresponding author.
E-mail addresses: thayer.kris@epa.gov (K.A. Thayer), angrish.michelle@epa.gov (M. Angrish), Arzuaga.Xabier@epa.gov (X. Arzuaga), Carlson.Laura@epa.gov
(L.M. Carlson), Davis.Allen@epa.gov (A. Davis), Dishaw.Laura@epa.gov (L. Dishaw), Druwe.Ingrid@epa.gov (I. Druwe), Gibbons.Catherine@epa.gov
(C. Gibbons), Glenn.Barbara@epa.gov (B. Glenn), Jones.Ryan@epa.gov (R. Jones), Kaiser.Jonathan-Phillip@epa.gov (J. Phillip Kaiser), Keshava.Channa@epa.gov
(C. Keshava), keshava.nagu@epa.gov (N. Keshava), kraft.andrew@epa.gov (A. Kraft), lizarraga.lucina@epa.gov (L. Lizarraga), persad.Amanda@epa.gov
(A. Persad), radke-farabaugh.elizabeth@epa.gov (E.G. Radke), rice.glenn@epa.gov (G. Rice), schulz.brittany@epa.gov (B. Schulz), Shaffer.rachel@epa.gov
(R.M. Shaffer), Shannon.Teresa@epa.gov (T. Shannon), Shapiro.Andy@epa.gov (A. Shapiro), Thacker.Samuel@epa.gov (S. Thacker), Vulimiri.Sury@epa.gov
(S.V. Vulimiri), Williams.Antony@epa.gov (A.J. Williams), Woodall.George@epa.gov (G. Woodall), Yost.Erin@epa.gov (E. Yost), robyn.blain@icf.com (R. Blain),
katie.duke@icf.com (K. Duke), Ali.Goldstone@icf.com (A.E. Goldstone), pamela.hartman@icf.com (P. Hartman), kevin.hobbie@icf.com (K. Hobbie), Ingle.
Brandall@epa.gov (B. Ingle), courtney.lemeris@icf.com (C. Lemeris), cynthia.lin@icf.com (C. Lin), alexander.lindahl@icf.com (A. Lindahl), kristen.mckinley@icf.
com (K. McKinley), parnian.soleymani@icf.com (P. Soleymani), nicole.vetter@icf.com (N. Vetter).

https://doi.org/10.1016/j.envint.2022.107468
Received 27 April 2022; Received in revised form 1 August 2022; Accepted 9 August 2022
Available online 13 August 2022
0160-4120/Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
K.A. Thayer et al. Environment International 169 (2022) 107468

exposure-level-only studies in humans, pharmacokinetic models, and absorption, distribution, metabolism, and
excretion (ADME). The availability of New Approach Methods (NAMs) evidence is also tracked (e.g., high
throughput, transcriptomic, in silico, etc.). Genotoxicity studies may be considered as PECO relevant or sup­
plemental material, depending on the topic and context of the review. Standard systematic review practices (e.g.,
two independent reviewers per record) and specialized software applications are used to search and screen the
literature and may include the use of machine learning software. Mammalian bioassay and epidemiological
studies that meet the PECO criteria after full-text review are briefly summarized using structured web-based
extraction forms with respect to study design and health system(s) assessed. Extracted data is available in
interactive visual formats and can be downloaded in open access formats. Methods for conducting study eval­
uation are also presented which is conducted on a case-by-case basis, depending on the usage of the SEM.

Background and how to use this template those developed by the Provisional Peer-Reviewed Toxicity Value
(PPRTV) Program. SEMs can also be prepared to explore the available
Systematic Evidence Maps (SEMs) are proving to be a particularly literature for an individual chemical or groups of chemicals of emerging
valuable analysis tool to inform the scope of complex human health interest. A companion publication describes the genesis and more spe­
assessments, like those conducted by the US Environmental Protection cific applications of SEMs within the IRIS Program (Thayer, 2022,
Agency (EPA)’s Integrated Risk Information System (IRIS) Program. The 10259559}. The purpose of this document is to disseminate a template
same workflows can be used to produce SEMs for other EPA Office of that presents typical methods used to develop an IRIS SEM and a range
Research and Development (ORD) human health assessments, such as of options to communicate the findings (results) of the SEM.

Fig. 1. Overview of the SEM steps, guiding questions and considerations, and associated location in this template.

2
K.A. Thayer et al. Environment International 169 (2022) 107468

Table 1
Example populations, exposures, comparators, and outcomes (PECO) criteria.
PECO element Description

Population Human: Any population and lifestage (occupational or general population, including children and other sensitive populations).
Animal: Nonhuman mammalian animal species (whole organism) of any lifestage (including fetal, early postnatal, adolescents and adults). [Note:
Full-text retrieval is recommended for studies of transgenic model systems. These studies typically provide mechanistic evidence tracked as “potentially relevant
supplemental material” but may present phenotypic information in wildtype animals that meets PECO criteria but is not mentioned in the abstract].
[Note: For both human and animal studies, any study that includes populations that address susceptibility should also be tagged with that supplemental content
tag.]
Exposures Relevant forms:
[chemical X] (CAS number)
Other forms of [chemical X] that readily dissociate (e.g., list any salts, etc.)
Metabolites of interest, including metabolites used to estimate exposures to [chemical X]
Occupations that may be considered surrogates of exposure
Human: Any exposure to [chemical X] [via [oral or inhalation] route[s] if applicable]. Studies will also be included if biomarkers of exposure are
evaluated (e.g., measured chemical or metabolite levels in tissues or bodily fluids) but the exposure route is unclear or likely from multiple routes.
Other exposure routes, such as those that are clearly dermal, will be tracked during title and abstract screening and tagged as “potentially relevant
supplemental material.”
[Note: Specify if certain exposure assessment matrices or methods will NOT be included. Also, although many epidemiology studies measure multiple chemicals,
they are not considered mixture studies (a type of supplemental content) as long as the study presents health outcome analyses specific to the chemical(s) of
interest. For most assessments that include hazard assessment, exposure does not have to be quantitated, e.g., low versus high is considered to meet PECO.]
Animal: Any exposure to [chemical X] via [oral or inhalation] route[s] of >1 day duration, or any duration assessing exposure during reproduction
or development. Studies involving exposures to mixtures will be included only if they include an experimental arm with exposure to [chemical X]
alone. Other exposure routes, including [dermal or injection], will be tracked during title and abstract as “potentially relevant supplemental
material.” [Note: Typically, IRIS assessments to develop chronic toxicity values consider acute non-developmental exposure studies (<1 day duration) as
supplemental content, but this can be adjusted depending on assessment needs. Also, the assessment-specific protocol should specify if certain exposures/study
designs will NOT be included, or if a minimum number of dose or concentration levels tested in experimental animal studies is indicated.]
Comparators [Note: PECO Human, Example A (general SEM): A comparison or referent population exposed to lower levels (or no exposure/exposure below detection limits),
element name can be adjusted or exposure for shorter periods of time, or cases versus controls, or a repeated measures design. However, worker surveillance studies are considered
toward SEM-specific aims, e. to meet PECO criteria even if no statistical analyses using a referent group is presented. Case reports or case series of > 3 people will be considered to
g., Comparisons or meet PECO criteria, while case reports describing findings in 1–3 people will be tracked as “potentially relevant supplemental material.”
Comparison group may be Human, Example B (targeted SEM to identify studies suitable for dose response): Studies reporting effect measures (e.g., relative risk,
more precise for SEMs standardized mortality ratio, beta coefficients) based on a comparison or referent population exposed to lower levels (or no exposure/exposure below
targeted toward identifying detection limits), or cases versus controls, or a repeated measures design.
dose response suitable [Notes: Studies based exclusively on duration of exposure analyses (i.e., longer versus shorter exposure duration) are not likely to be informative for SEMs
studies.] focused on identifying studies plausibly suitable for dose response.]
Animal: A concurrent control group exposed to vehicle-only treatment and/or untreated control (control could be a baseline measurement, e.g.,
acute toxicity studies of mortality, or a repeated measure design).
Outcomes All health outcomes (cancer and noncancer). In general, endpoints related to clinical diagnostic criteria, disease outcomes, biochemical,
histopathological examination, or other apical/phenotypic outcomes are considered to meet PECO criteria
[Note: Studies meeting PECO criteria may also contain supplemental mechanistic content that describes biological or chemical events associated with phenotypic
effects. When this occurs, these studies are also tagged as having supplemental mechanistic information. This typically happens during full-text review or doing
the literature inventory. Full-text retrieval is recommended for studies of transgenic model systems that meet E and C criteria because they may present
phenotypic information in wildtype animals that meet P and O criteria but is not reported in the abstract.]

This template is organized by sections to match how SEMs are often • IRIS Assessment Plan for Vanadium and Compounds, Oral (U.S. EPA,
presented in journal articles (i.e., a study structure of introduction, 2020b),
methods, results, and conclusions). Within each section there are de­ • IRIS Assessment Plan for Inorganic Mercury Salts (U.S. EPA, 2019b),
scriptions of the type of content that can be included. When possible, and.
template language is provided when the process being described is • Systematic Review Protocol for perfluorobutanoic acid (PFBA),
similar across different chemical SEMs (e.g., methods to screen studies perfluorohexanoic acid (PFHxA), perfluorohexanesulfonic acid
for inclusion, study evaluation methods, data extraction and manage­ (PFHxS), perfluorononanoic acid (PFNA), and perfluorodecanoic
ment). The template also presents example language to describe SEM acid (PFDA) IRIS Assessments (U.S. EPA, 2019c).
components that are optional or very in level of detail across SEMs,
depending on their decision-making context. For example, there is an Examples of introductory text developed for PPRTV assessments can
optional section on study evaluation methods and a range of options for be found on the PPRTV website, e.g.,
the level of detail used to describe study design and results. When
template text is not possible to develop because the nature of the content • Provisional Peer-Reviewed Toxicity Values for Trans-
is very chemical-specific, the user is referred to publicly accessible ex­ Crotonaldehyde (U.S. EPA, 2021),
amples. Fig. 1 provides an overview of the SEM steps, guiding questions • Provisional Peer-Reviewed Toxicity Value assessment for Glyci­
and considerations, and the associated location in the template. daldehyde (CASRN 765-34-4) (U.S. EPA, 2020h).
SEM template
Examples are also available of introductory text for SEMs published
1. Introduction in journal articles (Yost et al., 2021; Keshava et al., 2020; Yost et al.,
2019; Walker et al., 2018; Carlson et al., 2022; Radke et al., In Press).
The introduction should include a rationale on why the SEM is being The introduction may contain information on chemical and physical
developed and background on the topic. The level of detail varies properties. Within the IRIS and PPRTV Programs, this information is
depending on the context for preparing the SEM. Examples of intro­ typically presented as a table of values and chemical structures (when
ductory text developed to support IRIS problem formulation can be available) from the EPA CompTox Chemicals Dashboard (https://comp
found in published IRIS Assessment Plans (IAPs) and Systematic Review tox.epa.gov/dashboard), also referred to as the “Dashboard” (Williams
Protocols e.g., et al., 2021). [Note:
The physicochemical properties in the summary tables are based on

3
K.A. Thayer et al. Environment International 169 (2022) 107468

Table 2 Table 2 (continued )


Categories of potentially relevant supplemental material. Category (Tag) Description Typical Assessment Use
Category (Tag) Description Typical Assessment Use
• These data are used to inventory and
Pharmacokinetics Data Potentially Informative to Assessment Analyses estimate the amount prioritization.
Classical Classical PBPK and PK model absorbed (A), distributed Standard operating
pharmacokinetic (PK) Pharmacokinetic or studies are included in (D), metabolized (M), procedures for PBPK/
or physiologically Dosimetry Model Studies: the assessment and and/or excreted (E). PK model evaluation
based pharmacokinetic Classical PK or dosimetry evaluated for possible • ADME data can also be and the identification,
(PBPK) model studies modeling usually divides use in conducting collected from human organization, and
the body into just one or quantitative subjects who have had evaluation of ADME
two compartments, which extrapolations. PBPK/ environmental or studies are outlined in
are not specified by PK models are workplace exposures that An Umbrella Quality
physiology, where categorized as are not quantified or Assurance Project Plan
movement of a chemical supplemental material fully defined. (QAPP) for PBPK models
into, between, and out of with the expectation • ADME data, especially (U.S. EPA, 2018).
the compartments is that each one will be metabolism and tissue
quantified empirically by evaluated for partition coefficient
fitting model parameters to applicability to address information, can be
ADME (absorption, assessment generated using in vitro
distribution, metabolism, extrapolation needs and model systems. Although
and excretion) data. This technical conduct. in vitro data may not be
category is for papers that Specialized expertise is as definitive as in vivo
provide detailed required for their data, these studies should
descriptions of PK models evaluation. also be tracked as ADME.
but are not PBPK models. Standard operating For large evidence bases
procedures for PBPK/ it may be appropriate to
• The data are typically the PK model evaluation separately track the in
concentration time- and the identification, vitro ADME studies.
course in blood or plasma organization, and [Note: Studies describing
after oral and or intrave­ evaluation of ADME environmental fate and
nous exposure, but other studies are outlined in transport or metabolism in
exposure routes can be An Umbrella Quality bacteria or model systems
described. Assurance Project Plan that are not applicable to
Physiologically Based (QAPP) for PBPK models humans or animals should
Pharmacokinetic or (U.S. EPA, 2018). not be tagged.]
Mechanistic Dosimetry
Supplemental Evidence Potentially Informative to Assessment Analyses
Model Studies: PBPK
Mechanistic endpoints Studies that do not meet Prioritized studies of
models represent the body
PECO criteria but report mechanistic endpoints
as various compartments
measurements that inform are described in the
(e.g., liver, lung, slowly
the biological or chemical mechanistic synthesis
perfused tissue, richly
events associated with sections; subsets of the
perfused tissue) to quantify
phenotypic effects related most informative
the movement of chemicals
to a health outcome. studies may become
or particles into and out of
Experimental design may part of the units of
the body (compartments)
include in vitro, in vivo (by analysis. Mechanistic
by defined routes of
various routes of exposure; evidence can provide
exposure, metabolism, and
includes all transgenic support for the
elimination, and thereby
models), ex vivo, and relevance of animal
estimate concentrations in
in silico studies in effects to humans and
blood or target tissues.
mammalian and biological plausibility
nonmammalian model for evidence integration
• A defining characteristic
systems. Studies using New judgments [including
is that key parameters
Approach Methodologies MOA analyses, e.g.,
are determined from a
(NAMs; e.g., in vitro high using the MOA
substance’s
throughput testing framework in the US
physicochemical
strategies, read across EPA Cancer Guidelines
parameters (e.g., particle
applications) are also (2005)].
size and distribution,
categorized here. Studies
octanol-water partition
where the chemical is used
coefficient) and physio­
as a laboratory reagent (e.
logical parameters (e.g.,
g., as a chemical probe used
ventilation rate, tissue
to measure antibody
volumes).
response) generally should
Pharmacokinetic (ADME) Pharmacokinetic (ADME) ADME studies are not be tagged.
studies are primarily inventoried and Mechanistic evidence can
controlled experiments, prioritized for possible also help identify factors
where defined exposures inclusion in an ADME contributing to
usually occur by synthesis section on the susceptibility; these studies
intravenous, oral, chemical’s PK should also be tagged
inhalation, or dermal properties and for “susceptible populations.”
routes, and the conducting quantitative [Note: During screening,
concentration of particles, a adjustments or especially at the title and
chemical, or its metabolites extrapolations (e.g., abstract (TIAB) level, it may
in blood or serum, other animal-to-human). not be readily apparent for
body tissues, or excreta are Specialized expertise in studies that meet P, E, and C
then measured. PK is necessary for criteria if the endpoint(s) in

(continued on next page) (continued on next page)

4
K.A. Thayer et al. Environment International 169 (2022) 107468

Table 2 (continued ) Table 2 (continued )


Category (Tag) Description Typical Assessment Use Category (Tag) Description Typical Assessment Use

a study are best classified as the literature inventory


phenotypic or mechanistic rather than track as
with respect to the O supplemental.]
criteria. In these cases, the
Susceptible populations Studies that help to identify Provides information on
study should be screened as
potentially susceptible factors that might
“unclear” during TIAB
subgroups, including predispose sensitive
screening, and a
studies on the influence of populations or
determination made based
intrinsic factors such as sex, lifestages to a higher
on full-text review (in
lifestage, or genotype to risk of adverse health
consultation with a content
toxicity, as well as some effects following
expert as needed). Full-text
other factors (e.g., health exposure to the
retrieval is performed for
status). These are often co- chemical. This
studies of transgenic model
tagged with other information is
systems that meet E and C
supplemental material summarized during
criteria to determine if they
categories, such as evidence integration for
include phenotypic
mechanistic or ADME. each health effect and is
information in wildtype
Studies meeting PECO considered during dose-
animals that meet P and O
criteria that also address response, where it can
criteria that is not reported
susceptibility should be co- directly impact
in the abstract.]
tagged as supplemental. modeling decisions.
Non-PECO animal model Studies reporting outcomes Studies of non-PECO {Note: Susceptibility based
in animal models that meet animals, exposures, or on most extrinsic factors,
the outcome criteria but do durations can be such as increased risk for
not meet the population summarized to inform exposure due to residential
criteria in the PECO. evaluations of proximity to exposure
Depending on the consistency (e.g., across sources, is not considered
endpoints measured in species or routes or an indicator of susceptible
these studies, they can also durations), coherence, populations for the
provide mechanistic or adversity; subsets of purposes of IRIS
information (in these cases the most informative assessments.]
studies should also be studies may be included
Background Information Potentially Useful to Problem Formulation and Protocol
tagged “mechanistic in the unit of analysis.
Development (These studies fall outside the scope of IRIS assessment analyses)
endpoints”). These studies may also
Human exposure and Information regarding This information may
{Note: This categorization be used to inform
biomonitoring (no exposure monitoring be useful for developing
generally does not apply to evidence integration
health outcome) methods and reporting that exposure criteria for
studies that use species judgments of biological
are unrelated to health study evaluation or
with limited human health plausibility and/or
outcomes, but which refining problem
relevance (e.g., ecotoxicity- MOA analyses and thus
provide information on the formulation decisions.
focused studies are may be summarized as
following: methods for Notably, providing an
typically excluded]. part of the mechanistic
measuring human assessment of typical
Non-PECO route of Epidemiological or animal evidence synthesis.
exposure, biomonitoring (e. human exposures (e.g.,
exposure studies that use a non-PECO
g., detection of chemical in sources, levels) falls
route of exposure, e.g.,
blood, urine, hair), defining outside the scope of an
injection studies or dermal
exposure sources, or IRIS assessment.
studies if the dermal route
modeled estimates of
is not part of the exposure
exposure (e.g., in
criteria.
occupational settings).
{Note: This categorization
Studies that compare
generally does not apply to
exposure levels to a
epidemiological studies where
reference value, risk
the exposure route is unclear;
threshold or assessment
such studies are considered to
points of departure are also
meet PECO criteria if the
included in this category.
relevant route(s) of exposure
Studies related to
are plausible, with exposure
environmental fate and
being more thoroughly
transport are typically
evaluated at later steps.]
tagged as background
Non-PECO exposure For assessments that focus
materials unless otherwise
duration (optional) on chronic exposure, acute
described in the
exposure durations
assessment-specific
(defined as animal studies
protocol.
of less than 1 d) are
[Note: Assessment teams
generally considered
may want to subtag studies
supplemental. In rare cases
that describe or predict
and for very large evidence
exposure levels versus those
bases, short-term (i.e., less
that present exposure
than subchronic) exposure
assessment methods.]
durations could also be
categorized as Mixture study Mixture studies use Mixture studies are
supplemental. methods that do not allow tracked to help inform
{Note: Some assessment investigation of the health cumulative risk
teams may prefer to keep effects of exposure to the analyses, which may
these studies as PECO chemical of interest by itself provide useful context
relevant and summarize in (e.g., animal studies that for risk assessment but

(continued on next page)

5
K.A. Thayer et al. Environment International 169 (2022) 107468

Table 2 (continued ) curation, and validation of the content are ongoing. Any tables presented in
Category (Tag) Description Typical Assessment Use the SEM should be reviewed by chemists for obvious errors and to select the
most appropriate when multiple values were available.].
lack exposure to chemical fall outside the scope of
of interest alone or an IRIS assessment.
The introduction can also include a summary of existing health effect
epidemiology studies that values from federal, state, and international health agencies or other risk
do not evaluate associations assessment groups. Methods to conduct this summarization are pre­
of the chemical of interest sented in “Survey of Previous Assessments, Regulatory Reference
with relevant health
Values, Risk Thresholds or Assessment Based Points of Departure.”” The
outcome(s)).
[Note: Methods used to assess available values can be summarized graphically as a reference value
investigation of the exposure array or in a tabular format that includes derivation details; see example
by itself may not be clear figures and tables from the a naphthalene SEM (Yost et al., 2021), IRIS
from the abstract, in Assessment Plan for Vanadium and Compounds, Oral (U.S. EPA, 2020b),
particular for epidemiology
studies. When unclear, the
IRIS Assessment Plan for Inorganic Mercury Salts (U.S. EPA, 2019b),
study is advanced to full-text
review to determine 2. Methods
eligibility.]

Case reports or case series Human studies that present Tracking case studies The systematic review methods used to conduct the evidence map
an investigation of a single can facilitate awareness are outlined in the Office of Research and Development (ORD) Staff
exposed individual or group of potential human Handbook for Developing Integrated Risk Information System (IRIS)
of ≤ 3 subjects that describe health issues missed by
health outcomes after other types of studies
Assessments, referred to as the “IRIS Handbook” (U.S. EPA, 2020c).
exposure but lack a during problem [Note: The methods presented in this template reflect adjustments made based
comparison group (i.e., do formulation. on a November 2021 National Academy of Sciences, Engineering, and
not meet the “C” in the Medicine (NASEM) committee review of the November 2020 IRIS Handbook
PECO) and typically do not
(NASEM, 2021). SEM method changes made in response to NASEM focused
include reliable exposure
estimates. mostly on study evaluation, i.e., removing a reporting quality domain,
updating the sensitivity domain, and other minor changes to improve the
Other background Chemical-specific studies Although formal
information that may provide analyses on these
clarity of other domain considerations. Other changes were mostly editorial
introductory information general background or to improve the clarity of describing methods (e.g., updating phrasing to
on chemical and physical topics are not part of an match any changes made in revised handbook). The companion publication
properties (note: assessors IRIS assessment, this to this template (Thayer, 2022, 10259559} addresses a NASEM recom­
typically will separately information may be
mendation to clarify how SEMs are used in the IRIS assessment development
consult the EPA CompTox useful to protocol
Dashboard); sources, development (e.g., process.].
production, and use; and chemical properties
environmental occurrence information for 2.1. Specific aims
and fate. Additional evaluating PK or
groupings of information exposure in animal
may be determined on an studies). In addition, [Note: The specific aims for the SEM should be presented. Below are
assessment-specific basis brief summary a set of specific aims that can be adapted depending on the context for
and some assessments may overviews are typically conducting the SEM.].
decide to separately tag provided in the
different subsets of introductory materials.
• Survey core literature: Develop a SEM to identify epidemiological
information (e.g., tag
chemical properties studies (i.e., human) and toxicological (i.e., experimental animal) literature
separately from those on reporting health effects of exposure to [chemical X] as outlined by
environmental occurrence the Populations, Exposures, Comparators and Outcomes (PECO)
and fate).
criteria (Table 1).
Reference Materials • Identify supplemental content: Identify supplemental material as
Records with no original Records that do not contain Studies that are tracked outlined in Supplemental content (Table 2.) can include in vivo or ex
data original data, such as other for potential use in
vivo mechanistic, in vitro, or in silico studies; non-mammalian model
agency assessments, identifying missing
informative scientific studies, background systems; toxicokinetic and absorption, distribution, metabolism, and
literature reviews, information, or current excretion (ADME) studies; pharmacokinetic (PK) or physiologically-
editorials, or scientific opinions (e.g., based pharmacokinetic (PBPK) model studies; exposure character­
commentaries. hypothesized MOAs).
istics studies (no health outcome); data pertinent to identify sus­
Posters or conference Records that do not contain
abstracts sufficient documentation to ceptible populations (not otherwise identified by the PECO criteria);
support study evaluation mixture studies; routes of exposure not pertinent to the PECO; case
and data extraction. studies; references with no original data; and conference abstracts.
The availability of New Approach Methods (NAMs) evidence is also
tracked (e.g., high throughput, transcriptomic, in silico, etc.) pri­
information from a variety of sources, primarily from the EPA CompTox
marily through the mechanistic and non-mammalian model system
Chemicals Dashboard and PubChem. The data obtained from the EPA
categories. [Note: The example text provided in this template reflects a
CompTox Chemicals Dashboard are of varying quality and include both
SEM with no PECO restrictions placed on exposure duration, but for
experimental and predicted data [predicted data are annotated as such]. The
certain assessments acute duration studies may be considered supple­
data associated with the chemical substances in the CompTox Chemicals
mental and should be included in the list above].
Dashboard database have been compiled from public sources, databases and
• Provide visual overview: Create literature inventories to provide a
peer-reviewed literature and have varying levels of reliability and accuracy.
visual overview of studies that meet PECO criteria and are tagged as
Predicted data in particular have significant limitations in terms of the pre­
supplemental material.
dictions of properties for salts, inorganic and organometallic substances Links
• Evaluate studies (optional): Example A. Conduct study evaluation
to many external resources are provided in the Dashboard. Expansion,
of individual epidemiological and toxicological studies that meet

6
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 2. Overview of Integrated Risk Information System (IRIS) study evaluation process.(a) individual evaluation domains organized by evidence type, and (b)
individual evaluation domains judgements and definitions for overall ratings (i.e., domain and overall judgments are performed on an outcome–specific basis).

PECO criteria. [Note: Typically for SEMs, no study evaluation is con­ health outcome for hazard characterization, and identify priority
ducted for in vitro evidence, PBPK models, or other types of supplemental analyses of supplemental material to address key science issues,
material content, but this may be included if desired.] Example B: uncertainties in hazard characterization, and dose–response
Conduct study evaluation for epidemiological and toxicological analysis.
studies that are most suitable for dose response analysis based on
study design considerations.
• Summarize evidence base: Provide a narrative overview of the 2.2. Populations, exposures, comparators, and outcomes (PECO) criteria
evidence base that summarizes the available literature (including and supplemental material tagging
study evaluation observations, if conducted), identifying critical data
gaps and highlights key science issues. PECO criteria are used to focus the research question(s), search
• Include this specific aim when the SEM is being used to develop terms, and inclusion/exclusion criteria used in a SEM or systematic re­
a more focused PECO for subsequent assessment/full systematic view (Table 1). [Note: A set of SEM PECO criteria is presented in and
review (optional). Develop focus for subsequent assessment: presents alternative wording that can be used or adapted depending on the
Use the results of the SEM to develop refined PECO criteria for the context for conducting the SEM.] In addition to studies that meet the PECO
assessment, define the unit(s) of analysis at the level of endpoint or criteria and studies excluded as not relevant to the SEM, studies con­
taining supplemental material are inventoried during the literature

7
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 3. Study flow diagram (Example A: no machine learning software used and no/minimal references from gray literature found).

screening process. Table 2 presents major categories of supplemental are not expected to occur (U.S. EPA, 2002). Reference values are the
material. most common final output from the dose response assessment compo­
nent of the risk assessment paradigm set forth by the National Research
2.3. Literature search and screening processes Council (NRC, 2009, 1983) and are based on an observed or estimated
threshold for an effect, usually noncancer. Appendix A presents a list of
2.3.1. Survey of previous assessments, regulatory reference values, risk organizations that disseminate toxicity values that can be queried to
thresholds or assessment based points of departure conduct a survey. In addition to these sources, the ToxVal database on
Toxicity value is a broad term that encompasses reference values and the EPA CompTox Dashboard (https://comptox.epa.gov/dashboard/ch
cancer risk estimates (i.e., slope factors and unit risk estimates). The emical_lists/TOXVAL_V5) can be searched for reference values, risk es­
term reference value applies to values designed to provide a “bench­ timate values, and point of departure (PODs) as described in Appendix C
mark” or exposure limit below which adverse effects on human health (Williams et al., 2021). [Note: The data obtained from the EPA CompTox

8
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 4. Study flow diagram (Example B: no machine learning software used, numerous references from gray literature found, and SEM focused on identifying studies
potentially suitable for POD derivation). Studies can be tagged to multiple supplemental tags, therefore, total number of supplemental subtags is greater than the total
number of supplemental references.

Chemicals Dashboard, including the ToxVal database, are of varying quality. Table A1 and A2 from a naphthalene SEM (Yost et al., 2021). Values
In brief, ToxValDB collates publicly available toxicity dose–effect related without derivation details or that are derived from another agency’s
summary values typically used in chemical assessments. Many of the PODs value are summarized in a separate table from those values that have
presented in ToxVal are based on gray literature studies or assessments not derivation details available and are not shown in the toxicity value
available in databases such as PubMed and Web of Science (WoS). Although array. Tables and figures indicate the month and year the searches were
many of the resources included in Appendix A and below in “Searching Other conducted. A more detailed explanation of the development and evo­
Resources” are represented in ToxVal, they are also manually searched lution of the graphical toxicity value arrays used by the IRIS Program,
because most of the ToxVal entries have not undergone quality control to along with chemical-specific examples, can be found in reports (U.S.
ensure accuracy or completeness and might not include recent studies. EPA, 2013, 2009) and a peer-reviewed publication (Woodall et al., In
Searching ToxVal is considered optional and may perhaps be most useful Press).
when searching chemicals that are relatively data poor to identify unpub­
lished grey literature.]. 2.3.2. Literature search strategies
The available toxicity values can be summarized graphically as a
reference value array and/or in tabular format that includes derivation 2.3.2.1. Database search term development. Example A (EPA CompTox
details; see examples presented in Fig. 1, Fig. 2, Supplemental Materials Chemicals Dashboard as primary resource) [Note: Typically Example B is

9
K.A. Thayer et al. Environment International 169 (2022) 107468

used but use of the Chemicals Dashboard only can be a pragmatic option literature database search update is conducted several months prior to
when there is a need to simultaneously develop search strategies for large the planned dissemination of the SEM. Literature search update dates
numbers of chemicals, e.g., searches on hundreds of per-and polyfluoroalkyl are documented in the full details of the search strategy for each data­
substances (PFAS) compounds (Carlson et al., 2022).]. base (see Appendix B for example). Searching the other resources
The literature search focused on the chemical name (and synonyms, mentioned below is typically done once in the early phases of con­
trade names, and metabolites/degradants of interest) with no additional ducting the SEM.
limits. No language restrictions were applied. Chemical synonyms were The literature search should be updated to identify new literature
identified by searching the EPA CompTox Chemicals Dashboard (Wil­ published during assessment development and review. Studies identi­
liams et al., 2021; U.S. EPA, 2020a) and selecting those indicated as fied in literature search updates are generally screened according to
“valid” or “good.” The preferred chemical name (as presented in the both the SEM and refined PECO criteria, which allows the SEM to be
CompTox Chemicals Dashboard), CASRN, and synonyms are used to continually updated throughout assessment development; however,
identify existing toxicity values (See Appendix A) and shared with in­ depending on the size and scope of the assessment, the team may elect to
formation specialists who used these inputs to develop search strategies screen the literature search updates only according to refined PECO
tailored for each of the databases below because each database has its criteria. The last full literature search update is conducted several
own search architecture. Full details of the search strategy for each months prior to the planned release of the draft document for public
database are presented (see Appendix B for example presentation). comment.
Example B (SWIFT Review + EPA CompTox Chemicals Dashboard). Returned references (i.e., the number of “hits” from each electronic
The literature search focused on the chemical name (and synonyms, database or other literature source), including the results of any litera­
trade names, and metabolites/degradants of interest) with no additional ture search updates, are documented (see Appendix B for example) and
limits. No language restrictions applied. Chemical synonyms are iden­ the literature flow diagrams presented in Results. In addition to the
tified by first using the “Find Chemical Synonyms” feature in SWIFT databases listed below, a variety of other resources are subsequently
(Sciome Workbench for Interactive computer-Facilitated Text-mining) searched using customized processes (see “Other Resources”). [Note:
Review (Howard et al., 2016). In brief, this feature automatically creates Include mention of any special approaches to literature identification (e.g.,
a PubMed-formatted chemical search using (1) the common name for use of other assessments as starting points to identify evidence that meets the
the chemical as presented in the Tox21 chemical inventory list (U.S. PECO criteria.].
EPA, 2020i); (2) the Chemical Abstract Services Registry Number Accurate documentation of the search strategy is essential to the
(CASRN); (3) synonyms from the ChemIDPlus database, which currently systematic review process. Documentation of literature searches should
contains chemical names and synonyms for over 400,000 chemicals; and include the database(s) and date range covered by the search, search
(4) removal of ambiguous or short alphanumeric terms that could lead terms used and the filters (e.g., matching specific article types or
to false positives. This search is manually reviewed to ensure that any PubMed Medical Subject Headings [MeSH] terms; matching topic areas
synonyms listed in EPA’s CompTox Chemicals Dashboard (Williams in Web of Science) that were applied, and date(s) that the searches were
et al., 2021; U.S. EPA, 2020a) as “valid” or “good” are included. The performed (see an example template for documentation in Table 4–2).
preferred chemical name (as presented in the CompTox Chemicals
Dashboard), CASRN, and synonyms are used to identify existing toxicity • PubMed (National Library of Medicine).
values (See Appendix A) and shared with information specialists who • Web of Science (Thomson Reuters).
used these inputs The PubMed search created from SWIFT Review is • ProQuest.
shared with information specialists to develop search strategies tailored
for each of the databases below because each database has its own Example B (Database Searches and Subsequent Filtering of Retrieved
search architecture. Full details of the search strategy for each database References Using SWIFT Review).
are presented (see Appendix B for example presentation). The unique studies are imported into SWIFT Review software
When occupations are considered for surrogates for exposure (Howard et al., 2016) to identify those references most likely to be
(optional): When applicable, the SEM should include a methodological applicable to a human health risk assessment. In brief, SWIFT Review
description of how these searches were conducted. Typically, separate has pre-set literature search strategies (“filters”) developed by infor­
searches are conducted to identify pertinent occupational health studies mation specialists that can be applied to identify studies that are more
because these studies may not mention the chemical name in the title or likely to be useful for identifying human health content from those that
abstract. Resources such as prior assessments or reviews, information likely do not (e.g., environmental fate). The filters function like a typical
from the Occupational Safety and Health Administration (OSHA), or search strategy where studies are tagged as belonging to a certain filter if
information from the National Institute for Occupational Safety and the terms in the filter literature search strategy appear in title, abstract,
Health (NIOSH) can be used to identify the occupations. keyword or medical subject headings (MeSH) fields content. The applied
SWIFT Review filters focus on all lines of evidence used in human health
2.3.2.2. Database searches. Example A (Database Search Only) [Note: assessments: human, animal models for human health, and in vitro
Example A is typically only used when the number of records retrieved is studies. [Note: The lines of evidence filters can be tailored when needed, e.g.,
small and can be screened quickly. Example B allows filtering to the most human evidence only.] The details of the search strategies that underlie
records most likely pertinent to a human health assessment.] the filters are available online. Studies not retrieved using the search
The databases listed below are searched by an information specialist strategies are not considered further. Studies that include one or more of
in [insert month and year]. Retrieved references are imported into the the search terms in the title, abstract, keyword, or MeSH fields are
EPA’s Health and Environmental Research Online (HERO) database and exported as a RIS file for screening as described below. Application of
undergo a round of deduplication in HERO. Deduplication in HERO the SWIFT Review evidence stream filters reduced the number of studies
involves first determining whether a matching unique ID exists (e.g., for title and abstract screening from [insert pre-SWIFT Review number] to
PMID, WoSID, or DOI). If one matches one that already exists in HERO, [insert post-SWIFT Review number].
HERO will tag the existing reference instead of adding the reference [Note: SWIFT Review also has other filters that allow for more granular
again. Second, HERO checks if the same journal, volume, issue and page identification of human health literature, including health systems (cancer,
number are already in HERO. Third, HERO matches on the title, year, cardiovascular, developmental, endocrine, gastrointestinal, hematological
and first author. Title comparisons ignore punctuation and case. The and immune, hepatic, mortality, musculoskeletal, neurological, nutrition and
literature search is updated throughout SEM development. The last full metabolic, ocular and sensory, renal, reproductive, respiratory, skin and
connective tissue, ADME, PBPK, and epidemiologic quantitative analyses.

10
K.A. Thayer et al. Environment International 169 (2022) 107468

Other evidence stream filters pertinent to environmental assessments are also Centre) AcutoxBase. Data from the EU COSMOS database project
available, e.g., environmental fate, physicochemical properties, ecotoxicol­ have also been included in ToxValDB. Many of the PODs presented in
ogy, plant.]. ToxValDB are based on gray literature studies or assessments not
available in databases such as PubMed, WoS, etc. Although many of
2.3.2.3. Searching other resources. The literature search strategies the resources included in the “Other Sources Consulted” list are
described above are designed to be broad, but like any search strategy, represented in ToxValDB, they are also manually searched because
studies may be missed (e.g., cases where the specific chemical is not most of the ToxValDB entries have not undergone quality control to
mentioned in title, abstract, or keyword content; ability to capture ensure accuracy or completeness and might not include recent
“gray” literature that is not indexed in the databases listed above). Thus, studies. Searching ToxValDB can be helpful to provide an indication
in addition to the database searches, the sources below are used to of how much gray literature may be available for a chemical.
identify studies that may have been missed based on the database • European Chemicals Agency (ECHA) registration dossiers to identify
search. These sources are searched using customized processes described data submitted by registrants. Registration dossiers contain data on
in Appendix C. References that appear to meet the PECO criteria are substances such as hazardous properties, safe uses, classifications,
uploaded into the screening software, annotated with respect to source environmental fate, and ecotoxicological and toxicological infor­
of the record, and screened according to PECO as described below. mation. The amount of information provided for each substance
Searching of these sources is summarized to include the source type or varies and is obtained directly from companies’ REACH registrations.
name, the search string (when applicable), the URL (when available and ECHA does not give any guarantees or warranties regarding the
applicable), number of results, and number of unique references not quality and correctness of the published information. The informa­
otherwise identified from database searching. To identify unique refer­ tion in the portal is published ‘as provided’ by industry, and its ac­
ences, a citation for each identified study is generated in HERO and curacy has not been verified by ECHA (https://echa.europa.
verified that it is not already identified from the database searches (e.g., eu/information-on-chemicals/registered-substances).
PubMed, WoS, etc) prior to moving forward to screening. A training • EPA ChemView database (U.S. EPA, 2019a) to identify unpublished
guide for conducting the gray literature searches below is available in studies, information submitted to EPA under Toxic Substances Con­
the Health Assessment Workspace Collaborative (HAWC) project “SEM trol Act (TSCA) Section 4 (chemical testing results), Section 8(d)
Template Figures and Resources” (see “Tips for Searching Grey Litera­ (health and safety studies), Section 8(e) (substantial risk of injury to
ture” attachment). Some of the resources listed below are listed as health or the environment notices), and FYI (For Your Information,
optional, e.g., searches of gene expression databases may be more voluntary documents). Other databases accessible via ChemView
appropriate for data poor chemicals. [Note: Assessment creation on EPA include EPA’s High Production Volume (HPV) Challenge database
deployment of HAWC is currently limited to EPA staff and contractors. (https://iaspub.epa.gov/oppthpv/public_search.html_page) and the
However, a separate free and open source deployment, https://hawcproject. Toxic Release Inventory database.
org/ is available for hosting assessments not affiliated with EPA. This external • National Toxicology Program (NTP) Chemical Effects in Biological
website and content on it is not endorsed or supported by the US EPA; Systems (CEBS) database of study results and research projects (htt
however; it is nearly identical to EPA deployment and shares the same source ps://cebs.niehs.nih.gov/cebs/).
code.]. • The Organisation for Economic Cooperation and Development
(OECD) eChemPortal to retrieve results for OECD Screening Infor­
• Manual review (at the title level) of reference/citation list in studies mation DataSet (SIDS) and High Production Volume (HPV) Chem­
screened as PECO-relevant after full-text review. icals (https://www.echemportal.org/echemportal/).
• Review of the reference list from final or publicly available draft or • ECOTOX database (optional). Review of the list of references in the
finalized assessments (e.g., EPA IRIS [Integrated Risk Information ECOTOX database for the chemical(s) of interest (https://cfpub.epa.
System], ATSDR [Agency for Toxic Substances and Disease Registry] gov/ecotox/).
Toxicological Profile) or from published journal review specifically • References identified by technical consultants, during peer-review,
focused on human health. Reviews and assessments can be identified and during public comment periods (when applicable).
from the database search, the resources listed in Appendix A, or from • ToxCast or Tox21 high throughput screening information
the EPA CompTox Chemicals Dashboard ToxVal database (Williams (optional). The EPA CompTox Chemical Dashboard (Williams et al.,
et al., 2021; U.S. EPA, 2020a). 2021; U.S. EPA, 2020a) to retrieve a summary of any ToxCast or
• EPA CompTox Chemicals Dashboard ToxVal database Tox21 high throughput screening information. This data can be used
(optional.) Retrieval of references from EPA CompTox Chemicals to generate mechanistic insight, predict outcome using appropriate
Dashboard ToxVal database (Williams et al., 2021; U.S. EPA, 2020a) models, and potentially inform dose response modeling. Their
to identify studies or assessments that present Point of Departure importance for outcome prediction and dose–response modeling
(POD) information. ToxValDB collates publicly available toxicity depends on the context, size and quality of retrieved results and the
dose–effect related summary values typically used in risk assess­ lack of availability of other data typically used for these purposes.
ments. These include POD data collected from data sources within [Note: Searching for high throughput screening data is done on a case­
ACToR (Aggregated Computational Toxicology Resource) and Tox­ –by–case basis, largely dependent on the problem formulation (i.e., de­
RefDB, and no-observed and lowest-observed (adverse) effect levels cision context and acceptable level of confidence), size of the evidence
(NOEL, NOAEL, LOEL, LOAEL) data extracted from repeated dose base and need to track NAM evidence.].
toxicity studies submitted under REACH (Registration, Evaluation, • Comparative Toxicogenomics Database (CTD) (optional).
Authorisation and Restriction of Chemicals). Also included are Comparative Toxicogenomics Database (CTD), available at
reference dose and concentration values (RfDs and RfCs) from EPA’s https://ctdbase.org/. The CTD is a freely available public resource
Integrated Risk Information System (IRIS) and dose descriptors from that curates and interrelates chemical, gene/protein, phenotype,
EPA’s Provisional Peer-Reviewed Toxicity Values (PPRTV) docu­ disease, organism, and exposure data (Davis et al., 2021). A list of
ments. Acute toxicity information was extracted from a number of genes retrieved from the CTD can be informative, depending on its
different sources, including: OECD eChemPortal, ECHA (European size and context, and help generate mechanistic inferences or sup­
Chemicals Agency), NLM (National Library of Medicine) HSDB port mechanistic evidence generated by other studies. References
(Hazardous Substances Data Bank), ChemIDplus via EPA TEST that reported the chemical-gene interactions retrieved from the CTD
(Toxicity Estimation Software Tool), and the EU JRC (Joint Research can enrich the inventory of mechanistic references. [Note: Searching
for gene expression data is done on a case-by-case basis, largely

11
K.A. Thayer et al. Environment International 169 (2022) 107468

dependent on the problem formulation (i.e., decision context and translation tools or engagement with a native speaker.
acceptable level of confidence), size of the evidence base and need to Example B (use of machine learning to prioritize studies for TIAB
track NAM evidence.]. screening)
• Other gene expression databases (optional). Public data from The studies identified from the searches described above are im­
gene expression studies (Gene Expression Omnibus https://www.nc ported into SWIFT-Active Screener (https://www.sciome.com/swi
bi.nlm.nih.gov/geo/ and ArrayExpress https://www.ebi.ac.uk/arra ft-activescreener/) for TIAB screening. SWIFT-Active Screener is a
yexpress/). These data can be used to generate mechanistic evi­ web-based collaborative software application that utilizes active ma­
dence, predict adverse outcomes and derive toxicity values chine learning approaches to reduce the screening effort (Howard et al.,
depending on the design of gene expression studies that produced 2020). [Note: SWIFT–Active Screener is a machine learning based screening
these data. [Note: Searching for gene expression data is done on a case- application, however other screening software can be used, see
by-case basis, largely dependent on the on the problem formulation (i.e., https://systematicreviewtools.com/ for options]. TIAB screening is con­
decision context and acceptable level of confidence, size of the evidence ducted by two independent reviewers and any screening conflicts are
base and need to track NAM evidence.]. resolved by discussion between the primary screeners with consultation
• Backward and Forward Searches (optional). “Backward” searches by a third reviewer, if needed. A pilot screening effort is undertaken to
(to identify articles cited by included studies, reviews, or prior as­ confirm clarity of the screening criteria and adjust instructions. For ci­
sessments by other agencies) and “forward” searches (to identify tations with no abstract, articles are initially screened based on the
articles that cite those studies). [Note: This type of searching is done on following: title relevance (title should indicate clear relevance), and
a case-by-case basis depending on factors such as whether the PECO has a page length (articles two pages in length or less are assumed to be
targeted evidence stream or health outcome focus, extent of the evidence, conference reports, editorials, or letters). Eligibility status of non-
and use of other assessments to serve as a starting point. In general, the English studies is assessed using the same approach with online trans­
feasibility of conducting backward and forward searches is reduced when lation tools or engagement with a native speaker.
the PECO is broad, and the number of included studies is large. These The machine learning screening process is designed to prioritize
searches may be more appropriate to conduct when other assessments are references that appear to meet PECO-criteria or supplemental material
used as the starting point for a review and those other assessments were content for manual review (i.e., both types of references are screened as
either not conducted using systematic review methods.]. “include” for machine learning purposes). Screening continues until
SWIFT-Active Screener indicates that it is likely at least 95% of the
2.3.2.4. Non peer-reviewed data. It is possible that unpublished data relevant studies are identified, a percent identification often used to
directly relevant to the PECO may be identified during assessment evaluate the performance of machine learning applications and
development. In these instances, EPA will try to get permission to make considered comparable to human error rates (Bannach-Brown, 2018;
the data publicly available (e.g., in HERO); non peer-reviewed/ Howard et al., 2016; Cohen et al., 2006). Any studies with “partially
unpublished data that cannot be made publicly available are not used screened” status at the time of reaching the 95% threshold are then fully
in IRIS or PPRTV assessments. In addition, on rare occasions where screened. Studies identified as meeting PECO criteria, unclear, or sup­
unpublished data would be used to support key assessment decisions (e. plemental material during TIAB screening are then imported into Dis­
g., deriving a toxicity value), EPA may obtain external peer review if the tillerSR software (https://www.evidencepartners.com/products/dist
owners of the data are willing to have the study details and results made illersr-systematic-review-software/) either for conflict resolution or for
publicly accessible, or if an unpublished report is publicly accessible (or an additional round of more specific TIAB tagging (i.e., to separate
submitted to EPA in a non-confidential manner) (U.S. EPA, 2015). The studies meeting PECO criteria versus supplemental content and to tag
SEM may help identify studies to consider for external peer review in the evidence stream or specific type of supplemental content). [Note:
order to support an assessment analysis, but conducting that review DistillerSR is a commonly used screening application, but other screening
would be beyond the function of an SEM. software can be used, see https://systematicreviewtools.com/ for options]. In
Unpublished data from personal author communication can sup­ DistillerSR, TIAB screening is conducted by two independent reviewers
plement a peer-reviewed study as long as the information is made and any screening conflicts resolved by discussion between the primary
publicly available. If such ancillary information is acquired, it will be screeners with consultation by a third reviewer, if needed. Conflicts
documented in HAWC or HERO project page (depending on the nature between screeners in applying the supplemental tags, which primarily
of the information received). occur at the TIAB level, are resolved similarly, erring on the side of over-
tagging based on TIAB content.
2.3.3. Literature screening processes
2.3.3.2. Full-Text screening. Full-text references are sought through
2.3.3.1. Title and abstract (TIAB) screening. Example A (Full Manual EPA’s HERO database for studies screened as meeting PECO criteria or
Review). unclear based on the TIAB screening. Full-text screening occurs in Dis­
Studies identified from the database searches [and Swift Review, if tillerSR. Full-text copies of these records are independently assessed by
applicable] are imported into DistillerSR software (https://www. two screeners using a structured form in DistillerSR to confirm eligi­
evidencepartners.com/products/distillersr-systematic-review-softwa bility. Screening conflicts are resolved by discussion among the primary
re/) for screening [Note: DistillerSR is a commonly used screening appli­ screeners with consultation by a third reviewer or technical advisor (as
cation, but other screening software can be used, see needed to resolve any remaining disagreements). References that are not
https://systematicreviewtools.com/ for options]. TIAB screening is con­ able to be procured within 45 days of attempt are determined to be
ducted by two independent reviewers and any screening conflicts are unavailable. Rationales for excluding studies are documented, e.g.,
resolved by discussion between the primary screeners with consultation study did not meet SEM PECO, full-text not available. Approaches for
by a third reviewer, if needed. A pilot screening effort is undertaken to language translation include online translation tools or engagement of a
confirm clarity of the screening criteria and adjust instructions. For ci­ native speaker. Fee-based translation services for non-English studies
tations with no abstract, articles are initially screened based on the are typically reserved for studies that are anticipated as being useful for
following: title relevance (title should indicate clear relevance), and toxicity value derivation. When there are multiple publications using the
page length (articles two pages in length or less are assumed to be same or overlapping data, all publications are included. An example
conference reports, editorials, or letters). Eligibility status of non- screening form is available in Excel format (See Supplemental Excel file)
English studies is assessed using the same approach with online and access to the example screening form in DistillerSR is available upon

12
K.A. Thayer et al. Environment International 169 (2022) 107468

request (https://www.epa.gov/iris/forms/contact-us-about-iris) for re­ to ensure the supplemental material studies are categorized for easy
questors who have DistillerSR access. retrieval while conducting the assessment. Studies that meet the SEM
Other exclusions based on full-text content PECO are those most likely to be used to derive toxicity values and thus
In addition to failure to meet PECO criteria (described above), undergo subsequent individual study evaluation and data extraction. In
epidemiological and toxicological studies may be excluded at the full- contrast, it is often difficult to assess the impact of individual studies
text level due to critical reporting limitations. Reporting limitations tagged as supplemental material on the assessment conclusions during
can be identified during full-text screening but are more commonly literature screening. These studies could emerge as being critically
identified during subsequent phases of the assessment (e.g., literature important to the assessment and need to be evaluated and summarized
inventory, study evaluation). Regardless of when the limitation is at the individual study level (e.g., cancer MOA or ADME studies), help
identified, exclusions based on full-text content are documented at the characterize susceptible populations, or might be helpful to provide
level of full-text exclusions in literature flow diagrams with a rationale context (e.g., summarize current levels of exposure, provide hazard
of “critical reporting limitation.” Critical reporting information for evidence from routes or durations of exposure not pertinent to the SEM
different study types are summarized below. For each piece of infor­ PECO), or might not be cited at all in the assessment (e.g., individual
mation, if the information can be inferred (when not directly stated) for studies that contribute to a well-established scientific conclusion).
an exposure/endpoint combination, the study should be included.
Epidemiology studies 2.3.3.5. Literature flow diagrams. Results are summarized in a literature
flow diagram and interactive HAWC literature trees that show studies
• Sample size. “tagged” with appropriate category descriptors. i.e.., included, poten­
• Exposure characterization and/or measurement method. tially relevant supplemental material, excluded.
• Outcome ascertainment method.
• Study design.
2.4. Literature inventory/full data extraction
Animal studies
[Note: At a minimum, describing study methods and results (“data
• Species. extraction”) of studies that meet PECO criteria to create the literature in­
• Test article name. ventory should summarize basic features of study design and health system(s)
• Levels and duration of exposure. assessed. Depending on use of the SEM and available resources, data
• Route of exposure. extraction may include a qualitative description of results (i.e., indication on
whether effects or associations were reported but not at the level of capturing
Quantitative or qualitative (e.g., photomicrographs; author- effect size information) or a quantitative description of results that includes
reported lack of an effect on the outcome) results for at least one dose response and effect size information (“full data extraction”). Full data
endpoint of interest. extraction is not typically pursued for SEMs because it is time and resource
intensive. Full data extraction can be warranted depending on the purpose for
2.3.3.3. Multiple publications of the same data. When there are multiple conducting the SEM, e.g., journal publication or to determine whether new
publications using the same or overlapping data, all publications are evidence is likely to change an existing toxicity value. Template text covering
included, with one selected for use as the primary study; the others are these scenarios is presented below.].
considered as secondary publications with annotation in HAWC and
HERO indicating their relationship to the primary record during data 2.4.1. Literature inventory
extraction. For epidemiology studies, the primary publication is gener­ Human and animal studies that met PECO criteria after full-text re­
ally the one with the longest follow-up, the largest number of cases, and view are briefly summarized in DistillerSR using a structured data
/ or the most recent publication date. For animal studies, the primary extraction form [Note: DistillerSR is currently used to prepare literature
publication is typically the one with the longest duration of exposure, inventories, but similar capabilities are increasingly available in HAWC]. The
the largest sample size, and / or with the outcome(s) most informative to extraction fields used in these forms are available in Excel format (See
the PECO. For both epidemiology and animal studies, the assessments Supplemental Excel file) and access to the forms in DistillerSR is avail­
include relevant data from all publications of the study, although if the able upon request (https://www.epa.gov/iris/forms/contact-us-about
same data are reported in more than one study, the data is o­ -iris) for requestors who have DistillerSR access. These study sum­
nly extracted once. For corrections, retractions, and other companion maries are referred to as literature inventories and are exported from
documents to the included publications, a similar approach to annota­ DistillerSR in Excel format and imported into Tableau software (https
tion is taken and the most recently published data are incorporated into ://www.tableau.com/) or HAWC to create interactive visualizations
the assessments. (see Results for example formats). For non-English studies online
translation tools (e.g., Google translator), or engagement with a native
2.3.3.4. Supplemental material tagging. Supplemental material records speaker can be used to summarize studies at the level of literature in­
(Table 2) can be identified at both the TIAB or full-text levels. When ventory. Fee-based translation services for non-English studies are
tagged during title and abstract screening, it is not always clear whether typically reserved for studies considered potentially informative for dose
the chemical of interest is reported in the study (i.e., abstracts might not response based on an initial analysis of study design from the literature
describe all chemicals investigated). In such cases, studies are still tag­ inventory. Summaries are extracted into DistillerSR by one team mem­
ged with the expectation that additional screening would clarify if the ber and the extracted data are quality checked by at least one other team
study is pertinent. Conflicts between screeners in applying the supple­ member. Discrepancies are resolved by discussion between the primary
mental material tags, which typically occurs at the TIAB level, are extractors and additional consultation when needed.
resolved by discussion and consultation with a third reviewer (as Example A: Literature Inventory with Summarization of Study Design
needed), erring on the side of over-tagging at the TIAB level when the Only [Note: Compared to Examples B and C, this level of detail is the least
article content is relatively unclear. resource intensive to compile and is often used in preliminary assessment
It is important to emphasize that studies tagged as supplemental material (e.g., IAPs) as part of problem formulation to inform assessment
material are not necessarily excluded from consideration in any subse­ priorities and refine scope, identify data gaps, inform study evaluation
quent assessment analysis. The tagging structure in Table 2 is designed considerations.]
For animal studies, the following information is captured: chemical

13
K.A. Thayer et al. Environment International 169 (2022) 107468

assessed, study type (acute, short-term, subchronic, chronic, multigen­ for studies identified via the ToxVal (Williams et al., 2021) or the Eu­
erational, peripubertal, developmental), duration of treatment, route, ropean Chemical Agency (ECHA) databases, the study citation name
species, strain, sex, dose or exposure concentration levels tested, dose includes “(ToxVal)” or “(ECHA Summary)” to indicate the literature
units, health system, and specific endpoints assessed. [Note: Acute to inventory is based on limited report content.
short-term describes exposures of days to weeks and subchronic to chronic Because study evaluation and full quantitative extraction is time
describes exposure of months to years in animal toxicology studies. Specific intensive (often ranging from 30 min to 3 h or more per study for each
definitions of these study duration types may vary across different SEMs and activity), the literature inventory is used to prioritize studies for study
should be clearly stated in the SEM.]. evaluation and full quantitative extraction. To prioritize, the literature
For epidemiological studies, the following information is summa­ inventory inputs are analyzed with respect to the considerations below
rized: chemical assessed, population type (e.g., general population- to identify studies that plausibly could be considered suitable for iden­
adult, occupational, pregnant women, infants and children), study tifying a subchronic or chronic POD. Studies prioritized from this pro­
type (e.g., cross-sectional, cohort, case-control), sex, major route of cess undergo study evaluation (described below) and full data
exposure (if known), description of how exposure was assessed, health extraction. Studies not prioritized from this process are summarized at
system and specific outcomes assessed. the literature inventory level only and primarily used to identify data
For ToxVal and ECHA (optional). When the full-text report is not gaps, identify emerging health concerns, or to provide context when
available for studies identified via ToxVal (Williams et al., 2021) or interpreting findings from prioritized studies. [Note: Studies for which
ECHA, the study citation name includes “(ToxVal)” or “(ECHA Sum­ only ToxVal or ECHA summaries are available can advance to study eval­
mary)” to indicate the literature inventory is based on limited report uation or full extraction, depending on the amount of detail in the record and
content. goals of the SEM. However, these studies would not typically be used in an
Example B: Literature Inventory with Qualitative Summary of Study assessment because of lack of public access to the primary report.] Below are
Findings. [Note: Compared to Examples A and full data extraction, this level factors to consider when evaluating literature inventory results to pri­
of detail is intermediate with respect to resource intensity. This level of detail oritize studies for study evaluation and full quantitative extraction:
is the minimal amount needed for journal publication. If resources permit,
SEMs intended for journal publication should consider full-extraction (see • Studies with chronic exposure durations or including exposure dur­
below). This level of summarization may also be used in preliminary ing reproduction or development, are prioritized over studies with
assessment material (e.g., IAPs) as part of problem formulation to inform shorter-term exposure durations.
assessment priorities and refine scope, identify data gaps, inform study • Animal studies using a species that is considered a relevant human
evaluation considerations.]. surrogate.
For animal studies, the following information is captured: chemical • Studies with a broad exposure range and multiple exposure levels are
assessed, study type (acute, short-term, subchronic, chronic, multigen­ preferred to the extent that they can provide information about the
erational, peripubertal, developmental), duration of treatment, route, shape of the exposure–response relationship [see the EPA Benchmark
species, strain, sex, dose or concentration levels tested, dose units, Dose Technical Guidance, §2.1.1 (U.S. EPA, 2012)] and facilitate
health system and specific endpoints assessed, and a summary of find­ extrapolation to more relevant (generally lower) exposures. How­
ings at the health system level (null or no-observed-effect level/low- ever, single dose studies can be considered for toxicity value deri­
observed-effect level (NOAEL/LOAEL) based on author-reported statis­ vation if they test phenotypic health outcomes unexamined in
tical significance (when presented) with an indication of which specific multidose studies testing similar levels. For human studies, studies
endpoints were affected). [Note: Acute to short-term describes exposures of for which quantitative exposure measurements were available and
days to weeks and subchronic to chronic describes exposure of months to exposure–response results are presented in sufficient detail (e.g.,
years in animal toxicology studies. Specific definitions of these study duration standardized mortality rate or relative risks, numbers of cases/con­
types may vary across different SEMs and should be clearly stated in the trols, etc) are prioritized. Studies based exclusively on duration of
SEM.] For ToxVal and ECHA (optional). When the full-text report is exposure analyses (i.e., longer versus shorter exposure duration) are
not available for studies identified via the Toxicity Values (ToxVal) typically not considered suitable for dose response unless additional
(Williams et al., 2021) or the European Chemical Agency (ECHA) da­ information on exposure can be incorporated. Human studies include
tabases, the study citation name includes “(ToxVal)” or “(ECHA Sum­ experimental (e.g., controlled exposure) and observational studies.
mary)” to indicate the literature inventory is based on limited report • For epidemiological studies, studies that used biomarker measure­
content. ments in tissues or bodily fluids as the metric for exposure were only
For epidemiological studies, the following information is summa­ considered suitable for dose–response analysis if data or PBPK
rized: chemical assessed, population type (e.g., general population- models are available to extrapolate between the reported biomarker
adult, occupational, pregnant women, infants and children), study measurement and the level of exposure. For both animal and human
type (e.g., cross-sectional, cohort, case-control), sex, major route of studies, whether the nature of the outcomes/endpoints assessed were
exposure (if known), description of how exposure was assessed, specific interpretable with respect to potential adversity was considered.
endpoints assessed and a qualitative summary of findings at the health Typically, apical or clinical measures (“phenotypic”) are preferred
system level (null or an indication of any associations found). over other endpoints for dose response. However, “mechanistic”
Example C: Literature Inventory Used to Prioritize Studies for Subsequent endpoints can be useful in dose–response analyses when they can be
Study Evaluation and Full Data Extraction (Effect Size Level). reasonably established as predictive of, or strongly associated with,
For animal studies, the following information is captured: chemical phenotypic outcomes interpreted as adverse.
assessed, study type (acute, short-term, subchronic, chronic, multigen­ • High or medium confidence studies are highly preferred over low
erational, peripubertal, developmental), duration of treatment, route of confidence studies (see “Study Evaluation” below).
exposure, species, sex, and health system(s) assessed. [Note: Acute to
short-term describes exposures of days to weeks and subchronic to chronic 2.4.2. Full data extraction of study methods and results (optional)
describes exposure of months to years in animal toxicology studies. Specific [Note: Full data extraction includes the same data extraction elements as
definitions of these study duration types may vary across different SEMs and the literature inventory but is a more detailed summarization of study design
should be clearly stated in the SEM.] For studies in humans, the following and results compared to what is conducted in the literature inventory (e.g.,
information is summarized: chemical assessed, sex, population, study includes description of animal husbandry, effect size level presentation of
design, exposure measurement, and health system(s) assessed. For findings). For human studies the additional detail can include information on
ToxVal and ECHA (optional). When the full-text report is not available which covariates were included in the analysis. A variety of software

14
K.A. Thayer et al. Environment International 169 (2022) 107468

platforms can be used to conduct full data extraction and the specific informativeness (i.e., validity and sensitivity) of the results. The inde­
application should be specified, see https://systematicreviewtools.com/ for pendent reviewers use structured web-forms for study evaluation
options. The paragraph below has example sentences that describe sole use of housed within HAWC to record separate judgments for each domain and
the Health Assessment Workplace Collaborative (HAWC) or a combination the overall study for each outcome, to reach consensus between re­
of HAWC and DistillerSR.]. viewers, and when necessary, resolve differences by discussion between
Example A: Full extraction conducted completely in HAWC. Full data the reviewers or engagement of a tertiary reviewer. The web-forms are
extraction for prioritized studies is conducted using HAWC. General available in HAWC and can be customized to reflect the specific tool
instructions for data extraction in HAWC are available at https://haw being used for the study evaluation.
cprd.epa.gov/about/. An additional resource used to facilitate use of a For studies that examine more than one outcome of interest, the
consistent vocabulary to summarize endpoints assessed in animal evaluation process should be health outcome-specific, as the utility of a
studies is available in the HAWC project “SEM Template Figures and study may vary for the different outcomes. The same is true for epide­
Resources” [see the “Environmental Health Vocabulary (EHV) - a rec­ miology studies that assess exposure by different approaches. To cali­
ommended terminology for outcomes/endpoints”]. brate the assessment –specific considerations for study evaluations, a
Data extraction is performed by one member of the evaluation team pilot phase is conducted to assess and refine the evaluation process.
and quality checked by at least one other team member. The detailed Final study evaluations housed in HAWC are accessible from the
study data extracted in HAWC can be downloaded from EPA HAWC as a download page. Additional resources used to facilitate consistency of
Microsoft Excel file at [insert chemical specific HAWC project URL, see study evaluations are available in the HAWC project “SEM Template
“Download datasets”]. Figures and Resources” (see attachments for “Epidemiology Study
Example B: Full Extraction conducted in HAWC for Toxicology Studies Evaluation Core, Prompting, and Follow-up Questions,” “Example An­
and DistillerSR for Epidemiology Studies. Full data extraction for priori­ swers to the Animal Study Evaluation Domains,” and focused consid­
tized animal studies is conducted using HAWC. General instructions for erations for some types of animal toxicology studies described in “Study
data extraction in HAWC are available at https://hawcprd.epa. Evaluation Tips - Clinical Chemistry” and “Study Evaluation Tips -
gov/about/. An additional resource used to facilitate use of a consis­ Histopathology“).
tent vocabulary to summarize endpoints assessed in animal studies is During review, for each evaluation, domain reviewers reach a
available in the HAWC project “SEM Template Figures and Resources” consensus judgment of good, adequate, deficient, not reported, or critically
[see the “Environmental Health Vocabulary (EHV) - a recommended deficient (Fig. 2). These four categories are applied to each evaluation
terminology for outcomes/endpoints”]. For epidemiological studies, full domain for each study, as follows:
data extraction is conducted using DistillerSR. Access to the forms in
DistillerSR is available upon request (https://www.epa.gov/iris/fo • Good represents a judgment that the study was conducted appro­
rms/contact-us-about-iris) for users who have DistillerSR access. EPA priately in relation to the evaluation domain, and any minor de­
is in the process of updating HAWC’s epidemiology data extraction ficiencies noted would not be expected to influence the study results.
module and anticipates using HAWC instead of DistillerSR for full • Adequate indicates a judgment that methodological limitations
extraction by early 2023. related to the evaluation domain might occur, but those limitations
Data extraction is performed by one member of the evaluation team are unlikely to be severe or to notably impact the results.
and quality checked by at least one other team member. The detailed • Deficient denotes identified biases or deficiencies interpreted as likely
toxicology study data extracted in HAWC can be download from EPA to have had a notable impact on the results or that prevent inter­
HAWC to Excel at [insert chemical specific HAWC project URL, see pretation of the study findings.
“Download datasets”]. Detailed extraction of the epidemiology studies • Not reported indicates the information necessary to evaluate the
conducted in DistillerSR is available at [insert chemical specific HAWC domain question was not available in the study. Generally, this term
project URL, see “Attachments” at the bottom of the landing page]. carries the same functional interpretation as deficient for the pur­
poses of the study confidence classification (described below).
2.5. Study evaluation (optional) Depending on the number of unreported items and severity of other
limitations identified in the study, reaching out to the study authors
2.5.1. Human, animal, and in vitro evidence for this information might or might not be worthwhile.
The general approach for evaluating epidemiological, animal toxi­ • Critically deficient reflects a judgment that the study conduct relating
cological, and in vitro data is briefly described below. The approach for to the evaluation domain question introduced a serious flaw inter­
evaluating studies is conceptually the same, but the application specifics preted to be the primary driver of any observed effect(s) or makes the
differ based on evidence stream. Key concerns are potential bias (factors study uninterpretable. Studies with critically deficient judgments in
that affect the magnitude or direction of an effect in either direction) and any evaluation domain are almost always classified as overall
insensitivity (factors that limit the ability of a study to detect a true uninformative.
effect; low sensitivity is a bias toward the null when an effect exists). The
study evaluations are aimed at discerning the expected magnitude of any Once the evaluation domains have been rated, the identified
identified limitations (focusing on limitations that could substantively strengths and limitations are considered collectively to reach an overall
change a result), considering the expected direction of the bias. confidence classification of high, medium, or low confidence, or uninfor­
Examination of specific methodological features for each exposure- mative (Fig. 2) for each health outcome of interest. The classifications,
outcome/endpoint combination is accomplished by applying pre­ which reflect a consensus judgment between reviewers, are defined as
specified considerations to a set of domains. These domains differ for follows:
epidemiology, animal, and in vitro studies (Fig. 2). Core and prompting
questions for each domain are used to guide the reviewer to seek out and • High confidence: No notable deficiencies or concerns were identified;
think about relevant information pertaining to specific aspects of the the potential for bias is unlikely or minimal, and the study used
study. [Note: Any SEM-specific considerations should be described e.g., sensitive methodology. High confidence studies generally reflect
related to exposure or outcome assessment, confounders for epidemiology judgments of good across all or most evaluation domains.
studies. Example sentence: In addition to these base guides, specific consid­ • Medium confidence: Possible deficiencies or concerns were noted,
erations were developed for [describe SEM specific considerations]. but the limitations are unlikely to be notable. Generally, medium
At least two reviewers independently evaluate health effect studies confidence studies include adequate or good judgments across most
for study quality evaluation to identify characteristics that bear on the

15
K.A. Thayer et al. Environment International 169 (2022) 107468

domains, with the impact of any identified limitation not being The identification of studies during the screening process is sum­
judged as severe. marized in Fig. 3. The database searches yielded 4,388 references after
• Low confidence: Deficiencies or concerns are noted, and the potential duplicate removal in HERO. As described earlier, these studies were
for bias or inadequate sensitivity could have a significant impact on imported into SWIFT Review and literature search filters for human,
the study results or their interpretation. Typically, low confidence animal, and in vitro evidence applied. Application of these filters reduced
studies have a deficient evaluation for one or more domains, although the number of studies for TIAB screening to 1,918. During TIAB
some medium confidence studies might have a deficient rating in screening, 6 were included for full-text review, 13 were tagged as sup­
domain(s) considered to have less influence on the magnitude or plemental material, and 1,899 were excluded as not relevant to PECO.
direction of the outcome-specific results). Low confidence results are During full-text review, 3 studies were excluded as not relevant to PECO
given less weight compared to high or medium confidence results and 1 study was identified as supplemental. No references were identi­
during evidence synthesis and integration in an assessment. Studies fied from other sources or from a review of the reference lists for the 2
rated as medium or low confidence only because of sensitivity con­ included studies. Thus, only 2 studies (1 human and 1 animal) were
cerns about bias towards the null will be asterisked or otherwise considered to meet PECO criteria and included in the literature
noted because they might require additional consideration during inventory.
evidence synthesis. Effects observed in studies biased toward the null Example B: No Machine Learning Software Used, Numerous References
might actually increase confidence in the results, assuming the study Found from Gray Literature, and SEM Focused on Identifying Studies
is otherwise well conducted. Potentially Suitable for POD Derivation
• Uninformative: Serious flaw(s) make the study results unusable for The flow of studies during the screening process is summarized in
dose–response. Studies with critically deficient judgments in any Fig. 4. The database searches yielded 3,669 unique references after
evaluation domain are almost always classified as uninformative. duplicate removal. Application of the SWIFT Review filters (human,
Studies with multiple deficient judgments across domains also might animal/human health models, and in vitro) reduced the number of
be considered uninformative. Uninformative results are given the least studies for TIAB screening to 1,452. An additional 135 references found
weight during evidence synthesis and integration in an assessment. during searches of other sources (were added to the TIAB screen for a
Outcomes or endpoints judged as uninformative are not used as the total of 1,587 unique references (Appendix C, Table C1). After TIAB
basis for quantitative dose–response analysis. screening, 1,028 studies were excluded as not PECO relevant and
another 483 were tagged as supplemental material, leaving 76 studies
2.5.2. Pharmacokinetic models that advanced to full-text screening. During full text screening, 11 were
[Note: Pharmacokinetic models are not typically evaluated for SEMs, but excluded as not meeting PECO criteria and 5 were excluded due to an
if considered pertinent to the goals of the SEM then the following text can be inability to obtain the full-text, and another 14 were tagged as supple­
used] Any PBPK models identified in the SEM are evaluated using mental material.
methods described in the Quality Assurance Project Plan for PBPK Of the 46 studies considered PECO relevant, 28 were human studies
models (U.S. EPA, 2018), which are summarized below and presented in and 18 were animal studies. The human and animal studies were further
more detail in the IRIS Handbook. In practice, it has been found that inventoried as described earlier to identify those with study designs
many published models have errors that affect their predictions to most suitable for chronic POD derivation. Fifteen animal studies were
varying degrees; hence, an evaluation of a model is required before it not considered suitable for chronic POD derivation because they
can be used in an assessment (McLanahan et al., 2012). Thus, the models assessed acute (1 day) or short-term (<30 days) exposure duration,
identified as applicable to the assessment are evaluated as described leaving 31 studies (28 human, 3 animal) considered further for study
below. evaluation, suitability for dose response, and full extraction in HAWC.
Considerations for judging the suitability of a model are separated The human studies were considered potentially useful for hazard iden­
into two categories: scientific and technical. In summary, the scientific tification but not for identifying an oral or inhalation POD after analysis
criteria focus on whether the biology, chemistry, and other information of the exposure domain. Therefore, they were not fully extracted into
available for chemical mode(s) of action (MOA[s]) are appropriately HAWC.
represented by the model structure and equations. Significant to the Example text to highlight gray literature findings for key sources of animal
overall efficiency of this process, the scientific criteria can be judged by bioassay data:
reading the publication or report that describes the model, without Searches of the ToxVal database yielded 21 references, 6 of which
requiring an evaluation of the computer code. Preliminary technical were mammalian toxicity studies not identified from the database
criteria include the availability of the computer code and apparent search or other resources. These 6 studies were screened and inventoried
completeness of parameter listing and documentation. The in-depth in DistillerSR. Only 2 of these studies were developmental, reproductive,
technical and scientific criteria focus on the accurate implementation or repeated dose studies prioritized for study evaluation and full
of the conceptual model in the computational code, use of correct or extraction in HAWC.
biologically consistent parameters in the model, and reproducibility of The ECHA dossier had a total of 53 references, of which 39 were
model results reported in journal publications and other documents. study summaries and 14 were full study reports. Citation information
was not available for many of the ECHA references. Eight of the 53
3. Results references with citation information were studies not identified from
other resources and were assigned HERO ID numbers and screened in
3.1. Literature screening results DistillerSR. Twenty of the study summaries with citation information
matched references included in the database search or other resources.
[Note: Below are examples of presenting the literature screening results, The remaining 25 study summaries were cited as unnamed reports.
including example text that can be used when machine learning software is Three of these studies were determined to be duplicates of references
used, see https://systematicreviewtools.com/ for options. Links to editable already included in the literature inventory based on study year, animal
study flow diagrams in PowerPoint are also provided in the publicly accessible species, strain, sex, administered dose levels, and effect level de­
HAWC project “SEM Template Figures and Resources” (see the “Example scriptions. Of the remaining unnamed references, 5 described in vivo
Study Flow Figures”). Results may also be presented in interactive HAWC effects of inhalation or oral administration and were included in the
Literature Trees (see Example C, Fig. 7 below).]. literature using “unnamed report” as the citation information. The other
Example A: No Machine Learning Software Used and No/Minimal Ref­ unnamed reports describing mechanistic, basic toxicokinetics, and
erences from Gray Literature Found dermal administration studies were tagged to the appropriate

16
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 5. Study flow diagram (Example B PFAS 150 machine learning software used, numerous references from gray literature found). Studies can be tagged to
multiple supplemental tags, therefore, total number of supplemental subtags is greater than the total number of supplemental references.

17
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 6. PFAS 150 literature tree. Click here https://hawcprd.epa.gov/lit/assessment/100500085/references/visualization/ for interactive version.

18
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 7. Example A: Survey of PFAS 150 SEM human studies that met PECO criteria summarized by study design, population, and health systems assessed. This is a
screenshot image of the Tableau interactive dashboard for a literature inventory for a group of chemicals. The numbers in the heat map inset indicate the number of
studies that investigated a health system within a study design and population. If a study evaluated multiple health outcomes or populations, it is shown here multiple
times. The study details panel includes information on type of study, population, and exposure measurement. The dashboard is searchable by study design and health
system, but all also by chemical name or identifier (CASRN, DTXSID).

supplemental tags, with an additional “unnamed report” tag. (132 animal studies and 93 human studies), 94 studies were excluded,
One reference that was not identified in the journal database search and 430 studies were tagged as supplemental material. Of the 132 ani­
was found by searching EPA’s ChemView Database. This reference was mal studies, 35 were prioritized for study evaluation and full extraction
assigned a HERO ID and screened in DistillerSR according to PECO and because they were of 21 day or longer exposure duration, or exposures
supplemental material criteria. The reference reported an acute exper­ that occurred during reproduction or development. Literature search
iment (<24 h) and a short-term experiment that were both included in results are summarized graphically in Fig. 5 and Fig. 6. Chemical-
the literature inventory. specific literature trees are also available in HAWC [https://hawcprd.
Example C: Machine Learning Software Used, Numerous References from epa.gov/summary/assessment/https://doi.org/100500085/visuals/;
Gray Literature Found, HAWC Literature Tree (U.S. EPA, 2020e)] filter by visualization type “literature tagtree”).
The database searches for per- and polyfluoroalkyl substances
(PFAS) compounds yielded 40,623 references in HERO after duplicate 3.2. Characterizing human and animal evidence
removal (Fig. 5). Application of the SWIFT Review literature search
filters for human, animal, and in vitro evidence reduced the number of [Note: This section should describe the nature of the evidence base
studies for consideration to 10,397. The studies were screened in focusing on study design features, population or animal model system
Swift–Active Screener using predictive relevance, resulting in 3,690 assessed, health outcomes assessed, results (optional), and study evaluation
studies being manually screened to identify 1,075 studies that were (optional). The text should be concise. Web-based interactive visualizations
considered potentially PECO relevant or supplemental (“included” for that provide downloadable access to the summarized content are encouraged.
the purposes of machine learning) and 2,615 references that were The example text below includes presentations of optional content on study
excluded. After manually reviewing the 3,690 references, screening was evaluation and full data extraction. Six Tableau example visualizations are
stopped because Swift–Active Screener indicated that it was likely that presented in Figs. 8, 9, 10, 12, 13 and 14 to illustrate a variety of sum­
96% of the relevant studies were identified. This represented a signifi­ marized content for human and animal evidence:
cant reduction in screening effort (i.e., screening was stopped after ~
36% of the references were reviewed and 6,707 references were not • Human Evidence:
manually screened). o Study design only: Example A presented in Fig. 7 showing from a SEM
An additional 711 unique studies were identified from the other on 150 + PFAS chemicals.
sources searched, including 31 that came from reviewing the reference o Study design plus quantitative description of findings: Example C pre­
lists of studies considered PECO relevant after full-text review. These sented in Fig. 8 from an analysis of epidemiological studies identified in
711 studies were imported into DistillerSR for a total of 1,786 studies the SEM of 150 + PFAS chemicals.
screened at TIAB level. During TIAB screening in DistillerSR, 749 were o Study design plus qualitative description of findings: Example B pre­
included for full-text review, 819 were tagged as supplemental material, sented in Fig. 9 from the vanadium, oral IAP (2020). Note: Main
and 218 were excluded as not relevant to PECO. findings are visible via hover over features from within the Tableau
During full-text review, 225 studies were considered PECO relevant dashboard.

19
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 8. Example C. Survey of PFAS 150 human studies that met PECO criteria summarized by study design, population, and health systems assessed. This is a
screenshot image of the interactive dashboard. The numbers in the heat map inset indicate the number of studies that investigated a health system within a study
design and population. If a study evaluated multiple health outcomes or populations, it is shown here multiple times. The study details panel includes information on
type of study, population, exposure measurement, and findings with additional detail presented via website hover access.

• Animal Evidence: interactive dashboard with the, filterable by PFAS, health effect cate­
o Study design only: Example D presented in Fig. 11 from a SEM on 150 gory, outcome, population (adults, adults and children, children < 18
+ PFAS chemicals. years, pregnant women, occupational, or other), and study design is
o Study design plus qualitative description of findings (Example E): available at the hyperlink referenced in Fig. 7. The most commonly
Example E presented in Fig. 12 from the vanadium, oral IAP (2020). studied health effect categories were endocrine (22 studies), female
Note: Main findings are visible via hover over features from within the reproductive (21), developmental (21), metabolic (19), cardiovascular
Tableau dashboard.]. (17), and immune (13) effects. The other health effect categories
had<10 studies each. The plurality of studies were cross-sectional (50
studies), followed by cohorts (49), and case-control studies (19).
3.2.1. Human studies Example modified from vanadium, oral IAP (U.S. EPA, 2020b): A sur­
vey of study designs and health systems assessed in the human studies
3.2.1.1. Literature inventory. Example from SEM on 150+ PFAS chem­ that met PECO criteria and tabular summary of study design and find­
icals: A total of 118 publications of epidemiology studies were identified ings is provided in Fig. 9. Human studies identified in the literature
that included information on 10 different PFAS chemicals (Fig. 7). search included nine controlled trials that administered vanadyl sulfate
[Note: a presentation of studies with quantitative findings is presented in or sodium metavanadate directly to study participants. Of the controlled
Fig. 8]. The most frequently studied chemicals were per­ human trials, seven were conducted in diabetic patients for the purpose
fluoroundecanoic acid (86 studies), perfluoroheptanoic acid (26 of evaluating the therapeutic effects of vanadium supplementation, with
studies), perfluorooctanesulfonamide (23 studies), per­ treatment durations of 2–6 weeks (Afkhami-Arekani et al., 2008; Cusi
fluoroheptanesulfonate (18 studies), perfluorotridecanoic acid (17 et al., 2001; Goldfine et al., 2000; Boden et al., 1996; Halberstam et al.,
studies), and perfluorotetradecanoic acid (12 studies). The majority of 1996; Cohen et al., 1995; Goldfine et al., 1995); one evaluated effects of
the searched PFAS had zero epidemiology studies available. An vanadium supplementation on insulin sensitivity in healthy adults, with

20
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 9. Example B. survey of vanadium human studies that met PECO criteria summarized by study design, population, and health systems assessed. This is a
screenshot image of the interactive dashboard. The numbers in the heat map inset indicate the number of studies that investigated a health system within a study
design and population. If a study evaluated multiple health outcomes or populations, it is shown here multiple times. The study details panel includes information on
type of study, population, exposure measurement and main findings via website hover access.

a treatment duration of 7 days; and one evaluated effects of vanadium confounding was identified as well.
supplementation in weight training athletes, with a treatment duration
of 12 weeks (Fawcett et al., 1997). The literature search also identified 3.2.2. Animal studies
39 observational epidemiology studies, which evaluated the association
of health outcomes with total vanadium but the specific form of vana­ 3.2.2.1. Literature inventory. Example from SEM on 150+ PFAS chem­
dium was not determined. This included 37 studies (n = 13 case-control, icals: Of the 133 animal studies, a subset of 41 studies focused on 21 day
14 cross-sectional, and 10 cohort) in which vanadium exposure was or longer exposure durations (or were reproductive or developmental
evaluated using biomonitoring of blood (whole blood, plasma, or studies) for 16 PFAS chemicals (Fig. 11). These study designs were
serum), urine, hair, seminal plasma, cerebrospinal fluid, saliva, or nails, considered most suitable for identifying a subchronic or chronic POD
but in which the route of exposure was unclear; and two ecological and were prioritized for study evaluation and full data extraction in
studies that evaluated the association of human health outcomes with HAWC. The most commonly assessed health systems included the
vanadium levels in soil, drinking water, or food. reproductive (e.g., male and female reproductive organ weights,
[Note: If SEM includes study evaluation, then a paragraph should be fertility); developmental (e.g., pup weight and viability, lactation index,
added to describe to overall findings.]. skull malformation); urinary (e.g., kidney weight, organ function mea­
Example study evaluation summary modified from PFAS 150: Study surements such as blood urea nitrogen); immunological (e.g., spleen
evaluation results for the 26 available epidemiology studies of per­ and/or thymus weight; blood components such as monocytes, eosino­
fluoroheptanoic acid (PFHpA) are summarized in Fig. 10 and rationales phils, and lymphocytes); and hepatic (e.g., liver weight, enzyme activity,
are available in HAWC (for interactive graphic: https://hawcprd.epa. cholesterol, lipid metabolism, total bilirubin, nonneoplastic lesions such
gov/summary/visual/assessment/https://doi.org/100500085/PFAS- as hepatocellular hypertrophy). Fig. 12 present examples visualizations
150-Epidemiology-Study-Evaluations-for-PFHpA/). Twelve studies of the data extraction for some of the PFAS animal data. Access to the full
were medium confidence for at least one outcome, nine studies were low set of extracted results (~80 visualizations of extracted data from 16
confidence for all outcomes, and four studies were uninformative for all PFAS) is available at https://hawcprd.epa.gov/summary/assessment/
outcomes. While the majority of studies were generally well conducted, https://doi.org/100500085/visuals/.
with adequate and good ratings in most domains, most available studies
were deficient for study sensitivity. This was primarily due to low 3.2.2.2. Literature inventory with summary of findings (optional).
exposure levels with limited contrast in the study populations. The Example modified from vanadium, oral IAP (U.S. EPA, 2020b): A pre­
median exposure in most studies (16/26) was<0.1 ng/mL, with the liminary survey of study designs, species, vanadium compounds evalu­
highest median 0.4 ng/mL. Based on this poor sensitivity, null results in ated, and health effects evaluated in the animal studies that met PECO
this set of studies should not be interpreted as evidence of a lack of ef­ criteria is provided in Fig. 13. The animal studies evaluated exposure to
fect. In addition, none of the available studies were evaluated as good for ammonium metavanadate, sodium metavanadate, sodium orthovana­
confounding. Across studies, this was influenced by the uncertainty date, vanadyl sulfate, vanadium pentoxide, calcium orthovanadate, or
caused by potential for confounding across PFAS (due to moderate-high calcium pyrovanadate. Of these, vanadyl sulfate and sodium meta­
correlations), but in most studies, other potential for residual vanadate were the most frequently studied compounds. Two studies

21
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 10. Example A: summary of study evaluation for PFAS 150 SEM human studies of perfluoroheptanoic acid. Click here https://hawcprd.epa.gov/summary/visua
l/assessment/100500085/PFAS-150-Epidemiology-Study-Evaluations-for-PFHpA/ for interactive version.

reported that animals were exposed to “ammonium vanadate” (Susić reproductive or developmental studies that tested multiple dose levels
and Kentera, 1986) and “sodium vanadate” (Sun et al., 2014), which are provided in [Figs. 7, 8, and 9], respectively [Note: The referenced
were inferred to be ammonium metavanadate and sodium meta­ figures are presented in the vanadium, oral IAP and not the SEM template]. In
vanadate (respectively) based on the synonyms described above and are general, these study designs are preferred for toxicity value derivation
referred to accordingly here. Four studies reported that animals were over acute/short-term studies or studies that test a single dose level (U.S.
exposed to “vanadium” or “vanadate” but the specific chemical form EPA, 2002), although there may be circumstances where other study
was unclear. The majority of studies were conducted in rats and mice, designs are more suitable.
but data were also available in rabbits, cattle, goats, and sheep. Among
the 94 available animal studies, 23 included experiments in animal 3.2.2.3. Study evaluation (optional). Example from SEM on 150+ PFAS
models of diabetes that evaluated the therapeutic effects of vanadium chemicals: Study evaluations of the animal studies are summarized in
compounds on diabetic symptoms. Tabular summaries of the study de­ Fig. 14 and rationales for each domain and overall confidence ratings
signs and health effects evaluated in chronic, subchronic, and are available in HAWC (click here https://hawcprd.epa.gov/summar

22
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 11. Example D: Survey of PFAS 150 SEM animal studies that met PECO criteria by study design, species, and health systems assessed. This is a screenshot image
of the interactive dashboard that is filterable by health system, study design, PFAS name, CASRN, and DTXSID. The numbers in the heat map inset indicate the
number of studies that investigated a health system within a study design. If a study evaluated multiple health outcomes or presented several experiments, it is shown
here multiple times. The study details panel includes information on animal model, exposure duration, route of administration, dose level(s) tested, and presentation
of NOEL and LOEL information with additional detail presented via website hover access.

Fig. 12. Survey exposure-response array of liver weight findings among the 3 most studied PFAS included in the SEM with animal toxicology evidence (tri­
fluoroacetic acid (CAS 76–05-1), 6:2 fluorotelomer alcohol (CAS 647–42-7), 6:2 fuorotelomer methacrylate (CAS 17527–29-6)). Click here https://hawcprd.epa.gov/
summary/data-pivot/assessment/100500085/Main-Report_Hepatic_Weight_Oral/ (U.S. EPA, 2020f) for interactive version. All symbols reflect author-reported,
statistically significant changes only.

23
K.A. Thayer et al. Environment International 169 (2022) 107468

Fig. 13. Example E. survey of vanadium animal studies that met PECO criteria by study design, species, and health systems assessed. This is a screenshot image of the
interactive dashboard that is filterable by health system, study design, and specific vanadium compound. The numbers in the heat map inset indicate the number of
studies that investigated a health system within a study design. If a study evaluated multiple health outcomes or presented several experiments, it is shown here
multiple times. The study details panel includes information on animal model, exposure duration, route of administration, dose level(s) tested, and presentation of
NOEL and LOEL information with additional detail presented via website hover access.

Fig. 14. Study evaluation results for animal studies in the PFAS 150 SEM. Figure is presented as a screenshot for reference. Click here https://hawcprd.epa.
gov/summary/visual/100500560/ for interactive version (U.S. EPA, 2020d).

y/visual/100500560/ (U.S. EPA, 2020d) to see interactive graphic). The This figure shows the four risk-of-bias rating options. For each rating,
majority of the identified literature for PFAS were from ECHA sum­ the associated colored square and symbol and a definition of the rating
maries; however, these summaries typically lack details on the methods are provided. Good (metric) or High confidence (overall) is indicated by
and results, resulting in ratings of low confidence or uninformative for a dark green square with two plus signs. Adequate (metric) or Medium
specific endpoints or overall. For the most part, studies that were not confidence (overall) is indicated by a light green square with one plus
ECHA summaries were considered well conducted (medium or high sign. Deficient (metric) or Low confidence (overall) is indicated by a
overall confidence for the outcomes assessed (click here https://hawcpr yellow square with one minus sign. A yellow square with “NR” for not
d.epa.gov/summary/visual/100500561/ to see interactive graphic) (U. reported indicates there is insufficient information provided about the
S. EPA, 2020g). The exception was Feng et al. (2015), in which the metric. Critically deficient (metric) or Uninformative (overall) is indi­
Nafion was administered as Nafion membrane (which is comprised of cated by a dark red square with two minus signs. Color hatching with an
and releases fluoride) and the study’s internal fluoride measurements asterisk occurs where multiple judgements exist. A black solid triangle
indicated that the Nafion was not absorbed. Outcome-specific judge­ inside a square is used to indicate a result that biases away from null.
ments sometimes varied within a study usually based on the presenta­
tion of the results.

24
K.A. Thayer et al. Environment International 169 (2022) 107468

3.2.3. Mechanistic evidence


[Note: Mechanistic evidence identified from database searches is tagged
as supplemental content for SEMs. The information may be inventoried at a
high level describing the biological focus of the reference. In some cases, other
resources are searched (see “Other Resources” in Literature Search and
Screening Strategies) to identify high throughput screening information or
gene expression information that might be identified from database searches.
Searching for high throughput screening or gene expression data is done
on a case-by-case basis and is largely dependent on the problem
formulation (i.e., decision context and acceptable level of confidence),
size of the evidence base, and the need to track NAM evidence.].
Example modified from 1-Naphthol SEM.

3.2.3.1. Results from database search. The mechanistic studies tagged as


supplemental material were categorized by the biological focus of the
mechanistic information. Importantly, this categorization was typically
done based on TIAB content only. Full-text retrieval for supplemental
material was not done for the SEM. Access to the underlying references
is provided through the interactive version of Fig. 15.

3.2.3.2. ToxCast and Tox21 high throughput screening data (optional).


ToxCast and Tox21 high throughput screening data are available for
1–naphthol in the EPA CompTox Chemicals Dashboard (Fig. 16; click
here https://comptox.epa.gov/dashboard/dsstoxdb/results?search
=DTXSID6021793#bioactivity to see in the Dashboard) (Williams

Fig. 15. 1-Naphthol studies identified as mechanistic during literature


screening. Click here https://hawcprd.epa.gov/lit/assessment/100500049/refe
rences/visualization/ to view interactive version.

Fig. 17. Top interacting genes for 1–naphthol from the CTD.

Fig. 16. ToxCast activity summary for 1–naphthol.

25
K.A. Thayer et al. Environment International 169 (2022) 107468

et al., 2021). curation, Writing – review & editing. Glenn Rice: . Brittany Schulz:
Data curation, Writing – review & editing. Rachel M. Shaffer: Writing –
3.2.3.3. Comparative toxicogenomics database (optional). 1–Naphthol is original draft. Teresa Shannon: . Andrew Shapiro: Software. Shane
included in the Comparative Toxicogenomics Database (CTD). Below is Thacker: Software. Suryanarayana V. Vulimiri: Methodology,
a summary of the top interacting genes based on analysis of 57 studies Writing – review & editing. Antony J. Williams: Methodology, Writing
presented in the CTD (Fig. 17; click here http://ctdbase.org/detail.go? – review & editing. George Woodall: Methodology, Writing – review &
type=chem&acc=C029350 to see in the CTD). Note, these 57 studies editing. Erin Yost: Conceptualization, Methodology, Writing – original
were reviewed to identify any that were not otherwise retrieved from draft, Visualization. Robyn Blain: Methodology, Writing – review &
other sources (Appendix C). editing. Katherine Duke: Methodology, Writing – review & editing.
Alexandra E. Goldstone: Methodology, Writing – review & editing.
3.2.3.4. Gene expression omnibus, and array express (optional). No in Pam Hartman: Methodology, Writing – review & editing. Kevin Hob­
vivo transcriptomic references were identified for 1-naphthol from bie: Methodology, Writing – review & editing. Brandall Ingle: Meth­
Array Express (https://www.ebi.ac.uk/arrayexpress/search.html?que odology, Writing – review & editing. Courtney Lemeris: Methodology,
ry = 90–15-3). Two references were identified from Gene Expression Writing – review & editing. Cynthia Lin: Methodology, Writing – review
Omnibus (ID200092754 or 200060343), although neither was an in & editing. Alex Lindahl: Methodology, Writing – review & editing.
vivo study. One of these studies was not identified from other sources Kristen McKinley: . Parnian Soleymani: Methodology, Writing – re­
and was imported into DistillerSR for screening (Appendix D). view & editing. Nicole Vetter: Methodology, Writing – review &
editing.
4. Conclusions
Declaration of Competing Interest
The conclusion section of an SEM should include observations on the
extent and nature of the evidence and identification of key data gaps.
The authors declare that they have no known competing financial
When study evaluation is conducted, the conclusion section can also
interests or personal relationships that could have appeared to influence
indicate strengths and weaknesses in the evidence base. Examples of
the work reported in this paper.
SEM conclusion text can be found in Yost et al. (2021); Keshava et al.
(2020); Yost et al. (2019); Carlson et al. (2022); Radke et al. (In Press)
and Walker et al. (2018). Acknowledgements

CRediT authorship contribution statement The authors would like to thank the following individuals: Vicki Soto
and Dahnish Shams for document production support; and Barbara
Kristina A. Thayer: Conceptualization, Methodology, Writing – Soares, Casey Lindberg, Brittany Jacobs and Stan Barone for internal
original draft. Michelle Angrish: Conceptualization, Methodology, review. The authors would also like to thank staff at ICF for assistance in
Writing – review & editing. Xabier Arzuaga: Conceptualization, helping to evaluate and refine the workflows presented in the SEM.
Methodology, Writing – review & editing. Laura M. Carlson: Concep­ Funding and Disclaimer: Authors declare no financial conflict of
tualization, Methodology, Writing – original draft, Visualization. Allen interest. Funding for U.S. Environmental Protection Agency (EPA) au­
Davis: Methodology. Laura Dishaw: Methodology. Ingrid Druwe: thors was provided through employment with no additional external
Methodology. Catherine Gibbons: Methodology. Barbara Glenn: funding. Funding for ICF International authors was provided under EPA
Methodology. Ryan Jones: Software. J. Phillip Kaiser: Methodology. contract 68HERC19D0003. The views expressed are those of the authors
Channa Keshava: Methodology. Nagalakshmi Keshava: Methodol­ and do not necessarily represent the views or policies of the U.S. EPA.
ogy. Andrew Kraft: Conceptualization, Methodology, Writing – review Any mention of trade names, products, or services does not imply an
& editing. Lucina Lizarraga: Methodology. Amanda Persad: Method­ endorsement by the U.S. Government or the U.S. EPA. The EPA does not
ology. Elizabeth G. Radke: Conceptualization, Methodology, Data endorse any commercial products, services, or enterprises.

Appendix A. Survey of existing toxicity values

Table A1 lists websites which are searched for relevant human health reference values, along with indications of the results of the search. In
addition to these sources, the Toxicity Values (ToxVal) database on the EPA CompTox Chemicals Dashboard (https://comptox.epa.gov/dashboard/ch
emical_lists/TOXVAL_V5) is searched for both reference values and points of departure (POD)s as described in Appendix C. The ToxVal database is a
compilation of publicly available in vivo toxicity data, including toxicity values presented in a variety of assessments (Williams et al., 2021). [Note: Fill
in Search results column to indicate where results can be found (e.g., See Table 2); if no values were found, write “No values found”. Also note values without
derivation details should be summarized in a separate table from those values that have derivation details available.]
Background
Health-based reference values for noncancer effects are presented either in units of concentration (e.g., mg/L or mg/m3) or in terms of dose (e.g.,
mg/kg-day). Reference values generally are derived by applying uncertainty and adjustment factors to the POD. The POD is either an observed (e.g.,
NOAEL) or estimated (e.g., benchmark lower bound exposure/dose level that elicits an adverse effect observed in studies with human subjects or in
controlled animal experiments (U.S. EPA, 2002). The derivation methods and factors used in moving from a POD to a final reference value varies
according to the organization developing the values, often with consideration of how the resulting values will be applied. Most organizations develop a
human equivalent dose (HED) or concentration (HEC) in this process to account for dosimetric differences between humans and test species. Oral
reference values often are used as the basis for deriving standards for drinking water or acceptable levels in food.
Risk estimates are most often developed for cancer effects where the default assumption is that there is no level of exposure without some effect (i.
e., non-threshold effects) and estimates of risk related to exposure is part of the formulation (U.S. EPA, 2005). As noted previously, noncancer risks
have traditionally been managed by defining a level of exposure (i.e., reference value) below which there is a presumption of minimal health risk but
with no ability to estimate risk relative to any other exposure level. Approaches based on probability distributions to estimate distribution for
noncancer effect levels have also been developed more recently (IOMC ED, 2017; Chiu and Slob, 2015).

26
K.A. Thayer et al. Environment International 169 (2022) 107468

Assessment-based PODs or HECs/HEDs, along with uncertainty/adjustment factors, can also be used in a margin of exposure (MOE) approach to
determine exposure or consumption levels associated with acceptable level of risk. For example, risk evaluations performed under the Toxic Sub­
stances Control Act (TSCA, (Frank R. Lautenberg Chemical Safety for the 21st Century Act, 2016) apply an MOE approach as part of the risk decision
process on specific conditions of use for a subject chemical.

Table A1
Sources searched for existing human health reference values.
Source Search Results Query and/or link

ACGIH e.g., See Table 2. ACGIH. 2007. 2007 TLVs and BEIs: Based on documentation of the threshold limit values for chemical substances and physical agents
and biological exposure indices. Cincinnati, OH: American Conference of Governmental Industrial Hygienists.
AIHA e.g. No results found. AIHA. 2019. 2019 ERPG/WEEL Handbook. Fairfax, VA: American Industrial Hygiene Association. [Latest list of values.]
AIHA. 2002 (and updates). 2002 Emergency Response Planning Guidelines. Fairfax, VA: American Industrial Hygiene Association.
[Details used in deriving values.]
ATSDR e.g., Intermediate https://www.atsdr.cdc.gov/toxprofiles/index.asp
Inhalation MRL of 3 https://wwwn.cdc.gov/TSP/MRLS/mrlsListing.aspx
mg/m3
EPA CompTox https://comptox.epa.gov/dashboard
Chemicals
Dashboard
CT DEEP https://eregulations.ct.gov/eRegsPortal/Browse/getDocument?guid=%7b00D6A654-0300-CC47-9B95-397D2AD21304%7d
DFG https://onlinelibrary.wiley.com/doi/https://doi.org/10.1002/9783527826889.oth
OW (US EPA) https://www.epa.gov/sdwa/drinking-water-contaminant-human-health-effects-information
EPA/NRC AEGL https://www.epa.gov/aegl/access-acute-exposure-guideline-levels-aegls-values#chemicals
European https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32017L0164&from = EN
Commission
Health Canada https://www.canada.ca/en/services/health/publications/healthy-living.html
https://publications.gc.ca/site/archivee-archived.html?url=https://publications.gc.
ca/collections/collection_2012/sc-hc/H128-1–11-638-eng.pdf
https://publications.gc.ca/site/archivee-archived.html?url=https://publications.gc.ca/collections/Collection/H46-2–96-194E.pdf
HEAST https://epa-heast.ornl.gov/heast.php
https://cfpub.epa.gov/ncea/risk/hhra/recordisplay.cfm?deid= 2877
HSA https://www.hsa.ie/eng/publications_and_forms/publications/codes_of_practice/chemical_agents_cop_2020.pdf
IDEM https://www.in.gov/idem/toxic/2343.htm
ID DEQ https://adminrules.idaho.gov/rules/current/58/580101.pdf
IFA https://limitvalue.ifa.dguv.de/WebForm_gw2.aspx
IRIS https://www.epa.gov/iris/
ITER https://iter.tera.org/database.htm
JSOH https://www.sanei.or.jp/?mode= view&cid = 328
MassDEP https://www.mass.gov/service-details/massdep-ambient-air-toxics-guidelines
MDH https://www.health.state.mn.us/communities/environment/risk/guidance/air/table.html
https://www.health.state.mn.us/communities/environment/risk/guidance/gw/table.html
MI EGLE https://www.michigan.gov/documents/deq/deq-rrd-chem-CleanupCriteriaTSD_527410_7.pdf
NATICH https://nepis.epa.gov/Exe/ZyPDF.cgi/2000NS7S.PDF?Dockey= 2000NS7S.PDF
NC DEQ https://files.nc.gov/ncdeq/Air%20Quality/rules/rules/D1104.pdf
NDEP https://ndep.nv.gov/resources/risk-assessment-and-toxicology-basic-comparison-levels
NIOSH https://www.cdc.gov/niosh/npg/npgdcas.html
https://www.cdc.gov/niosh/pubs/criteria_date_desc_nopubnumbers.html
https://www.cdc.gov/niosh/idlh/intridl4.html
NJ DEP https://www.state.nj.us/dep/aqpp/downloads/risk/ToxAll2020.pdf
NY DEC https://www.dec.ny.gov/docs/remediation_hudson_pdf/techsuppdoc.pdf
OAQPS (US EPA) https://www.epa.gov/fera/dose response-assessment-assessing-health-risks-associated-exposure-hazardous-air-pollutants
OEHHA https://www.oehha.ca.gov/tcdb/index.asp
Ontario MOL https://www.labour.gov.on.ca/english/hs/pubs/oel_table.php
OPP (US EPA) https://iaspub.epa.gov/apex/pesticides/f?p= chemicalsearch:1
OR DEQ https://www.oregon.gov/deq/FilterDocs/airtox-abc.pdf
OSHA https://www.osha.gov/chemicaldata/
PAC Database https://edms.energy.gov/pac/Search
PPRTV https://www.epa.gov/pprtv/provisional-peer-reviewed-toxicity-values-pprtvs-assessments
Publications https://legisquebec.gouv.qc.ca/en/showdoc/cr/S-2.1,%20r.%2013?csi_scan_9222d36c6a354dc6=BO9xyrMZ +
Quebec 270UP3j0MGuOD0kZjgFAAAAXrM3HA==&bcsi_scan_filename = S-2.1,%20r.%2013&bcsi_scan_9222d36c6a354dc6 =
KXzmpPueuN0L1AjnJOB1Zerr85YMAAAAyhrPTg==&bcsi_scan_filename = S-2.1,%20r.%2013
RI DEM https://www.dem.ri.gov/programs/benviron/air/pdf/airtoxgl.pdf
RIVM https://www.rivm.nl/bibliotheek/rapporten/711701092.pdf
https://www.rivm.nl/bibliotheek/rapporten/609021044.pdf
https://www.rivm.nl/bibliotheek/rapporten/711701025.pdf
Safe Work https://www.safeworkaustralia.gov.au/exposure-standards#exposure-standards-in-australia
Australia
SWCAA https://www.swcleanair.org
TCEQ https://www.tceq.texas.gov/toxicology/dsd/final
https://www.tceq.texas.gov/remediation/trrp/trrppcls.html
USAPHC https://phc.amedd.army.mil/topics/envirohealth/hrasm/Pages/TG230.aspx
VT DEC https://dec.vermont.gov/sites/dec/files/aqc/laws-regs/documents/AQCD%20Regulations%20ADOPTED_Dec132018.
pdf#page=127
WAC https://apps.leg.wa.gov/WAC/default.aspx?cite=173–460-150
Worksafe https://worksafe.govt.nz/topic-and-industry/work-related-health/monitoring/exposure-standards-and-biological-exposure-indices/
WHO https://www.who.int/ipcs/publications/ehc/en/
https://www.who.int/water_sanitation_health/publications/drinking-water-quality-guidelines-4-including-1st-addendum/en/

ACGIH = American Conference of Governmental Industrial Hygienists; AEGL = Acute Exposure Guideline Levels; ATSDR = Agency for Toxic Substances and Disease
Registry; BEI = biological exposure index; CT DEEP = Connecticut Department of Energy & Environmental Protection; DFG = Deutsche Forschungsgemeinschaft, German
27
K.A. Thayer et al. Environment International 169 (2022) 107468

Research Foundation; EPA = Environmental Protection Agency; HEAST = Health Effects Assessment Summary Tables; HSA = Health and Safety Authority; IDEM = Indiana
Department of Environmental Management; ID DEQ = Idaho Department of Environmental Quality; IFA = Institut für Arbeitsschutz, The Institute for Occupational Safety
and Health; IRIS = Integrated Risk Information System; ITER = International Toxicity Estimates for Risk Assessment; JSOH = Japan Society for Occupational Health;
MassDEP = Massachusetts Department of Environmental Protection; MDH = Minnesota Department of Health; MI EGLE = Michigan Environment, Great Lakes & Energy;
MOL = Ministry of Labour; NATICH = National Air Toxics Information Clearinghouse; NC DEQ = North Carolina Department of Environmental Quality; NDEP = Nevada
Division of Environmental Protection; NIOSH = National Institute for Occupational Safety and Health; NJ DEP = New Jersey Department of Environmental Protection; NRC
= National Research Council; NY DEC = New York State Department of Environmental Conservation; OAQPS = Office of Air Quality Planning and Standards; OEHHA =
California Office of Environmental Health Hazard Assessment; OPP = Office of Pesticide Programs; OR DEQ = Oregon Department of Environmental Quality; OSHA =
Occupational Safety and Health Administration; OW = Office of Water; PAC = Protective Action Criteria; PPRTV = Provisional Peer-Reviewed Toxicity Value; RI DEM =
Rhode Island Department of Environmental Management; RIVM = Rijksinstituut voor Volksgezondheid en Milieu, The Netherlands Institute for Public Health and the Envi­
ronment; SWCAA = Southwest Clean Air Association; TCEQ = Texas Commission on Environmental Quality; TLV = threshold limit value; USAPHC = United States Army
Public Health Center; VT DEC = Vermont Department of Environmental Conservation; WAC = Washington Administrative Code; WHO = World Health Organization.

Appendix B. Literature search strategies

See Table B1.

Table B1
Example presentation for results of literature search strategy.
Database Termsa Search Date and
Results

Web of Science (TS=“chloroform” OR TS=“1,1,1-trichloromethane” OR TS=“chloroforme” OR TS=“trichloromethane”) AND PY=(2000–2018) 12/3/2018


NOT (SU=“physics” OR SU=“plant sciences” OR SU=“energy fuels” OR SU=“instruments instrumentation” OR SU=“computer Number of hits: 2,344
science” OR SU=“legal medicine” OR SU=“metallurgy metallurgical engineering” OR SU=“mechanics” OR SU=“education
educational research” OR SU=“acoustics” OR SU=“geochemistry geophysics” OR SU=“mathematics” OR SU=“forestry” OR 5/31/2019
SU=“automation control systems” OR SU=“mining mineral processing” OR SU=“construction building technology” OR
SU=“astronomy astrophysics” OR SU=“archaeology” OR SU=“operations research management science” OR SU=“anthropology” Number of hits: 30
OR SU=“sport sciences” OR SU=“art” OR SU=“paleontology” OR SU=“telecommunications” OR SU=“chemistry” OR SU=“polymer
science” OR SU=“engineering” OR SU=“environmental sciences ecology” OR SU=“food science technology” OR SU=“science 7/10/2019
technology other topics” OR SU=“biotechnology applied microbiology” OR SU=“agriculture” OR SU=“spectroscopy” OR Number of hits: 7
SU=“crystallography” OR SU=“integrative complementary medicine” OR SU=“water resources” OR SU=“nutrition dietetics” OR
SU=“life sciences biomedicine other topics” OR SU=“parasitology” OR SU=“thermodynamics” OR SU=“optics” OR 4/30/2020
SU=“biophysics” OR SU=“tropical medicine” OR SU=“veterinary sciences” OR SU=“research experimental medicine” OR
SU=“marine freshwater biology” OR SU=“meteorology atmospheric sciences” OR SU=“geology” OR SU=“electrochemistry” OR Number of hits: 65
SU=“general internal medicine” OR SU=“dentistry oral surgery medicine” OR SU=“entomology” OR SU=“nuclear science
technology” OR SU=“infectious diseases” OR SU=“fisheries” OR SU=“oceanography” OR SU=“anesthesiology” OR SU=“zoology” 2/25/2021
OR SU=“virology” OR SU=“radiology nuclear medicine medical imaging” OR SU=“medical laboratory technology” OR Number of hits: 99
SU=“mycology” OR SU=“surgery” OR SU=“biodiversity conservation” OR SU=“obstetrics gynecology” OR SU=“evolutionary
biology” OR SU=“psychiatry” OR SU=“remote sensing” OR SU=“pediatrics” OR SU=“mineralogy” OR SU=“transplantation” OR Total hits: 2,545
SU=“microscopy” OR SU=“rheumatology” OR SU=“geriatrics gerontology” OR SU=“orthopedics” OR SU=“materials science”)

PubMed ((((“chloroform”[MeSH Terms] OR “1,1,1-trichloromethane”[All Fields]) OR “chloroforme”[All Fields]) OR “trichloromethane”[All 12/3/2018


Fields]) OR “67-66-3′′ [EC/RN Number]) AND (”2000/01/01′′ [PDAT]: “3000′′ [PDAT])
Number of hits: 2,527

5/31/2019
Number of hits: 68

7/10/2019
Number of hits: 4

4/30/2020
Number of hits: 56

2/25/2021
Number of hits: 57

Total hits: 2,712


ToxNetb @AND+@OR+(chloroform + chloroforme + trichloromethane+@TERM+@rn + 67 + 66 + 3)+@RANGE + yr + 2000 + 12/3/2018
2018+@NOT+@org + pubmed + pubdart+“nih + reporter”
Number of hits: 365

5/31/2019
Number of hits: 0

7/10/2019
Number of hits: 0

Total hits: 365


Merged Total after duplicate removal in HERO 5,003
Reference Set

Initial search and update history: 12/3/2018 (4,620 unique records identified); 5/31/2019 (98 unique records identified); 7/10/2019 (11 unique records identified);
4/30/2020 (119 unique records identified); 2/25/2021 (155 unique records identified); total of 5,003.
a
Publication year terms were updated for each search date. Other elements of the search strings did not change.
b
As of December 2019, the ToxNet database has shut down and no longer exists as a searchable information venue and its content was moved to other National
Library of Medicine products.
28
K.A. Thayer et al. Environment International 169 (2022) 107468

Appendix C. Process for searching and collecting evidence from selected other resources

An additional resource used to facilitate consistency in searching selected other sources and retrieving results is available in the HAWC project
“SEM Template Figures and Resources”, Gray Literature Training slides.
Review of reference lists from existing assessments (final or publicly available draft), journal reviews articles and studies considered
relevant to PECO based on full-text screening.
Review of the citation reference lists is typically done manually because they are not available in a file format (e.g., RIS) that permits uploading into
screening software applications. Manual review entails scanning the title, study summary, or study details as presented in the resource for those that

Table C1
Summary table of search results for other sources (1-naphthol example).
Sourcea Source address Search terms Search Total References not
date unique otherwise
number of identified that
results were screened
retrieved in DistillerSR

Review of reference lists NA NA NA 67 65


studies considered
relevant to PECO based
on full-text screening.
Review of reference lists NA NA NA 3 0
from existing
assessments (final or
publicly available draft)
or journal review
articles that focused on
human health
EPA CompTox (Chemicals https://comptox.epa.gov/dashboard/dsstoxdb/results? 90–15-3 (Results from Human 12/10/ 21 5
Dashboard (ToxVal) abbreviation= TOXVAL_V5&search = Health: POD, Toxicity Value, 2019
DTXSID6021793#toxicity-values Lethality Effect Level)
EPA CompTox Chemicals https://comptox.epa.gov/dashboard/dsstoxdb/results? 90–15-3 9/19/ 1 1
Dashboard (ToxCast or search= DTXSID6021793 2019
Tox21 high throughput
screening information)
ECHA https://echa.europa. 90–15-3 10/8/ 53 30
eu/information-on-chemicals/registered-substances 2019
EPA ChemView https://chemview.epa.gov/chemview?tf= 0&ch = 90–15- 90–15-3 9/19/ 3 1
3&su = 2–5-6–7-37574985&as = 3–10-9–8&ac = 1–15- 2019
16–6378999&ma = 4–11-1981377&tds = 0&tdl = 10&tas1 =
1&tas2 = asc&tas3 = undefined&tss=
High Production Volume https://ofmpub.epa.gov/oppthpv/quicksearch.display? 90–15-3 9/19/ 4 4
Challenge Database pChem= 101850 2019
(HPVIS)
NTP CEBS https://manticore.niehs.nih.gov/cebssearch/search?q= 90–15-3 9/19/ 0 0
90–15-3 2019
OECD eChemPortal https://www.echemportal.org/echemportal/substance-search 90–15-3 9/19/ 0 0
2019
ECOTOX Database https://cfpub.epa.gov/ecotox/search.cfm 90–15-3 9/19/ 4 3
2019
Comparative https://ctdbase.org/ 90–15-3 12/9/ 57 30
Toxicogenomics 2019
Database (CTD)
ArrayExpress https://www.ebi.ac.uk/arrayexpress/ 90–15-3 & “naphthol” 12/9/ 1 1
2019
Gene Expression Omnibus https://www.ncbi.nlm.nih.gov/geo/ (90–15-3[rn] OR “1- 12/9/ 2 1
Naphthol”[tw] OR “Naphthalen- 2019
1-ol”[tw] OR “1-
Naphthalenol”[tw] OR “1-
naphthalenol”[tw]) AND
(“Expression profiling by RT-
PCR”[Filter] OR “Expression
profiling by MPSS”[Filter] OR
“Expression profiling by
SAGE”[Filter] OR “Expression
profiling by SNP array”[Filter]
OR “Expression profiling by
array”[Filter] OR “Expression
profiling by genome tiling
array”[Filter] OR “Expression
profiling by high throughput
sequencing”[Filter] OR “Protein
profiling by Mass Spec”[Filter]
OR “Protein profiling by protein
array”[Filter]).
a
PECO = Populations, Exposures, Comparators, and Outcomes; NA = not applicable; POD = point of departure; ECHA = European Chemicals Agency; NTP CEBS =
National Toxicology Program Chemical Effects in Biological Systems; OECD = Organisation for Economic Co-operation and Development.

29
K.A. Thayer et al. Environment International 169 (2022) 107468

appear to meet the PECO criteria. Any references that are not identified from the other sources are formatted in a RIS file format, imported into
DistillerSR, annotated with respect to source, and screened as outlined in the methods section. For tracking assessments or reviews, the name of the
source citation and the number of references imported into DistillerSR are noted. The reference list of any study included in the literature inventory are
reviewed manually to identify titles that appeared relevant to the PECO criteria. These citations are tracked in a spreadsheet, compared against the
literature base to determine if they are unique to the project, and then added to DistillerSR to be screened at the title and abstract stage for PECO
relevance.
EPA CompTox Chemicals Dashboard (ToxVal)
ToxVal is searched in the EPA CompTox Chemicals Dashboard (https://comptox.epa.gov/dashboard), and data available from the Hazard tab is
exported (Williams et al., 2021). Using the human health POD summary file, citations are identified that apply to human health PODs. A citation for
each referenced study is generated in HERO and verified that it is not already identified from the database search (or searches of “other sources
consulted”) prior to moving forward to screening in DistillerSR. Full texts are retrieved where possible; if full texts are not available, data from the
ToxVal dashboard are entered and the citation annotated accordingly for Tableau and HAWC visualizations by adding “(ToxVal)” to the citation.
European Chemicals Agency (ECHA)
A search of the ECHA registered substances database (https://echa.europa.eu/information-on-chemicals/registered-substances) is conducted
using the CASRN. The registration dossier associated with the CASRN is retrieved by navigating to and clicking the eye-shaped view icon displayed in
the chemical summary panel. The general information page and all subpages included under the Toxicological Information tab are downloaded in PDF
(Portable Document Format), including all nested reports having unique URLs. Any pages indicated as a toxicological or endpoint summary are
omitted. In addition, data is extracted from each dossier page and used to populate an Excel tracking sheet. An example of a tracking sheet is available
in the Gray Literature Training slides in the HAWC project “SEM Template Figures and Resources”. Extracted fields include data from the general
information page regarding the registration type and publication dates, and on a typical study summary page the primary fields reported in the
administrative data, data source, methods, and results sections. Each study summary results in more than one row in the tracking sheet if more than
one data source or effect level is reported.
At this stage, each study summary is reviewed for inclusion based on PECO criteria. Study summaries labeled “read across” (if any) are screened in
DistillerSR and considered supplemental material. When a study summary considers relevant reported data from a study or lab report, a citation for the
full study is generated in HERO and verified that it is not already identified from the database search (or searches of “other sources consulted”) prior to
moving forward to screening. When citation information is not available and a full text cannot be retrieved, the generated PDF is used as the full text
for screening and extraction and the citation annotated accordingly for Tableau and HAWC visualizations by adding “(ECHA Summary)” to the
citation.
EPA ChemView
The EPA ChemView database (https://chemview.epa.gov/chemview) (U.S. EPA, 2019a) is searched using the chemical CASRN. The prepopulated
CASRN match and the “Information Submitted to EPA” output option filter are selected before generating results. If results are available, the icon
under the “Data Submitted to EPA” column is selected, and the following entries are included:

• High Production Volume Challenge Database (HPVIS).


• Human Health studies (Substantial Risk Reports).
• Monitoring (Includes environmental, occupational, and general entries).
• TSCA Section 4 (Chemical testing results).
• TSCA Section 8(d) (Health and safety studies).
• TSCA Section 8(e) (Substantial risk).
• FYI (Voluntary documents).

All references for ecotoxicology and physical and chemical property entries are excluded. When results are available, extractors navigate into each
record until a substantial risk report link is identified and saved as a PDF file. If the report cannot be saved, due to file corruption or broken links, the
record is not retrieved and is identified as “unable to obtain record.” Most substantial risk reports contain multiple document IDs, so citations are
derived by concatenating the unique report numbers (OTS; 8EHD Num; DCN; TSCATS RefID; and CIS) associated with each document along with the
typical author organization, year, and title. Once a citation is generated, the study moves forward to DistillerSR where it is screened according to PECO
and supplemental material criteria.
National toxicology Program (NTP) chemical effects in biological systems
This database is searched using the chemical CASRN (https://manticore.niehs.nih.gov/cebssearch). All non-NTP data are excluded using the “NTP
Data Only” filter. Data tables for reports undergoing peer review are also searched for studies that have not been finalized (https://ntp.niehs.nih.
gov/data/tables/index.html) based on a manual review of chemical names.
OECD eChemPortal
The OECD eChemPortal (https://www.echemportal.org/echemportal/substance-search) is searched using the chemical CASRN. Only database
entries from the following sources are included and entries from all other databases are excluded in the search. Final assessment reports and other
relevant SIDS reports embedded in the links are captured and saved as PDF files.

• OECD HPV.
• OECD SIDS International Uniform Chemical Information Database (IUCLID).

ECOTOX database
EPA’s ECOTOX Knowledgebase (https://cfpub.epa.gov/ecotox/search.cfm) is searched using the CASRN. Results are refined to terrestrial
mammalian studies by selecting the terrestrial tab at the top of the search page and sorting the results by species group.
EPA CompTox chemical dashboard search to retrieve ToxCast or Tox21 high throughput screening summary information
Version 3.0.9 of the CompTox Chemicals Dashboard (https://comptox.epa.gov/dashboard) is accessed for high throughput screening (HTS) data
by searching the Dashboard by CASRN (Williams et al., 2021). Next, the “Bioactivity” section is selected and the availability of ToxCast/Tox21 HTS
data for active and inactive assays is examined in the “TOXCAST: Summary” tab. If active assays are reported, the figure is copied for presentation in

30
K.A. Thayer et al. Environment International 169 (2022) 107468

the systematic evidence map. This figure presents (i) scatterplot of scaled assay responses vs. AC50 values for each active assay endpoint, and (ii)
cytotoxicity limit as a vertical line. More detailed information on the results of ToxCast and Tox21 assays are available in the CompTox Chemicals
Dashboard section “ToxCast/Tox21,” which includes chemical analysis data, dose response data and model fits, and “flags” assigned by an automated
analysis, which might suggest false positivity/negativity or indicate other anomalies in the data. This information is not summarized further for the
purposes of the systematic evidence map, which is focused on identifying the extent of available evidence.
Comparative toxicogenomics database
This CTD database (https://ctdbase.org/) is searched using the chemical CASRN in the “keyword search” with pulldown menu set to “Chemicals.”
The query results in the “Gene Interactions” tab provide the list genes or proteins reported in the published references as interacting with the query
chemical. Human and rodent genes/proteins interacting with a query chemical are identified and their numbers are provided in the systematic ev­
idence map. If information is available, a figure presenting the interacting genes available in the “Basics” tab is copied to the systematic evidence map.
When many interacting genes are identified it may be appropriate to limit to the top 10 or 20 to represent genes whose interactions with query
chemical are supported by most available references. Details on interaction types and degrees are provided at https://ctdbase.org/help/ixnQueryHe
lp.jsp. The top scoring pathway relevant to the identified interacting genes is provided in the systematic evidence map. The reference list of studies
reporting gene/protein interactions with the query chemical are compared to existing references in DistillerSR. Unique references are added to
DistillerSR and screened according to PECO and supplemental material criteria.
Gene expression omnibus, and ArrayExpress
Public repositories of omics data Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) and ArrayExpress (https://www.ebi.ac.
uk/arrayexpress/) are queried to identify available gene expression datasets relevant to the exposure by the chemical of interest. The GEO is queried
using the custom search string shown below, in which the letters “XX” are replaced by the chemical name as represented in the PubMed search strategy
from Appendix A. Filters are applied to select human, mouse and rat datasets. Note, the filter options might not be available in the GEO website if no
human, mouse, or rat datasets exist. Retrieved studies are reviewed and the studies that report gene expression data for chemical exposures of interest
are reported in the systematic evidence map as follows: Series accession number (GSExxxxx); title of study, species, route of exposure, platform and
tissue type.
GEO Search string
(“XX”[MeSH Terms] OR XX[All Fields]) AND (“Expression profiling by RT-PCR”[Filter] OR “Expression profiling by MPSS”[Filter] OR “Expression
profiling by SAGE”[Filter] OR “Expression profiling by SNP array”[Filter] OR “Expression profiling by array”[Filter] OR “Expression profiling by
genome tiling array”[Filter] OR “Expression profiling by high throughput sequencing”[Filter] OR “Protein profiling by Mass Spec”[Filter] OR
“Protein profiling by protein array”[Filter])
The ArrayExpress (https://www.ebi.ac.uk/arrayexpress/) repository is queried using chemical name as a keyword and filtered to limit datasets to
“Homo sapiens,” “Rattus norvegicus” and “Mus musculus” species. All studies reporting RNA-seq, transcription profiling, proteomic profiling or
translation profiling data are reported in the systematic evidence map.

Appendix D. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.envint.2022.107468.

References Feng, M., Qu, R., Habteselassie, M., Wu, J., Yang, S., Sun, P., Huang, Q., Wang, Z., 2015.
Hepatic transcriptome responses in mice (Mus musculus) exposed to the nafion
membrane and its combustion products. PLoS ONE 10, e0128591. https://doi.org/
Afkhami-Arekani, M., Karimi, M., Mohammadi, S.M., Nourani, F., 2008. Effect of sodium
10.1371/journal.pone.0128591.
metavanadate supplementation on lipid and glucose metabolism biomarkers in type
Frank R. Lautenberg Chemical Safety for the 21st Century Act, ,. Public Law 114-182 -
E diabetic patients. Malaysian J. Nutr. 14, 113–119.
Frank R. Lautenberg Chemical Safety for the 21st Century Act, (2016). https://www.
Bannach-Brown, A, ., P.,rzybyła, P.,., T.,homas, J.,., R.,ice, A.,.S.C., Ananiadou. S., L.,iao,
epa.gov/assessing-and-managing-chemicals-under-tsca/frank-r-lautenberg-
J.,., M.,acleod, M.,.R., 2018. The use of text-mining and machine learning algorithms
chemical-safety-21st-century-act.
in systematic reviews: reducing workload in preclinical biomedical sciences and
Goldfine, A.B., Patti, M.E., Zuberi, L., Goldstein, B.J., Leblanc, R., Landaker, E.J.,
reducing human screening error. Cold Spring Harbor Protocols 0: 1-26. http://dx.
Jiang, Z.Y., Willsky, G.R., Kahn, C.R., 2000. Metabolic effects of vanadyl sulfate in
doi.org/10.1101/255760.
humans with non-insulin-dependent diabetes mellitus: in vivo and in vitro studies.
Boden, G., Chen, X., Ruiz, J., 1996. Effects of vanadyl sulphate on carbohydrate and lipid
Metabolism 49, 400–410. https://doi.org/10.1016/S0026-0495(00)90418-9.
metabolism in patients with non-insulin dependent diabetes mellitus. Metabolism
Goldfine, A.B., Simonson, D.C., Folli, F., Patti, M.E., Kahn, C.R., 1995. Metabolic effects
45, 1130–1135.
of sodium metavanadate in humans with insulin-dependent and noninsulin-
Chiu, W.A., Slob, W., 2015. A unified probabilistic framework for dose-response
dependent diabetes mellitus in vivo and in vitro studies. J. Clin. Endocrinol. Metab.
assessment of human health effects [Review]. Environ. Health Perspect. 123,
80, 3311–3320. https://doi.org/10.1210/jcem.80.11.7593444.
1241–1254. https://doi.org/10.1289/ehp.1409385.
Halberstam, M., Cohen, N., Shlimovich, P., Rossetti, L., Shamoon, H., 1996. Oral vanadyl
Cohen, A.M., Hersh, W.R., Peterson, K., Yen, P.Y., 2006. Reducing workload in
sulfate improves insulin sensitivity in NIDDM but not in obese nondiabetic subjects.
systematic review preparation using automated citation classification. J. Am. Med.
Diabetes 45, 659–666. https://doi.org/10.2337/diab.45.5.659.
Inform. Assoc. 13, 206–219. https://doi.org/10.1197/jamia.M1929.
Howard, B.E., Phillips, J., Miller, K., Tandon, A., Mav, D., Shah, M.R., Holmgren, S.,
Cohen, N., Halberstam, M., Shlimovich, P., Chang, C.J., Shamoon, H., Rossetti, L., 1995.
Pelch, K.E., Walker, V., Rooney, A.A., Macleod, M., Shah, R.R., Thayer, K., 2016.
Oral vanadyl sulfate improves hepatic and peripheral insulin sensitivity in patients
SWIFT-Review: a text-mining workbench for systematic review. Syst. Rev. 5, 87.
with non-insulin-dependent diabetes mellitus. J. Clin. Invest. 95, 2501–2509.
https://doi.org/10.1186/s13643-016-0263-z.
https://doi.org/10.1172/JCI117951.
Howard, B.E., Phillips, J., Tandon, A., Maharana, A., Elmore, R., Mav, D., Sedykh, A.,
Cusi, K., Cukier, S., Defronzo, R.A., Torres, M., Puchulu, F.M., Redondo, J.C., 2001.
Thayer, K., Merrick, B.A., Walker, V., Rooney, A., Shah, R.R., 2020. SWIFT-Active
Vanadyl sulfate improves hepatic and muscle insulin sensitivity in type 2 diabetes.
Screener: Accelerated document screening through active learning and integrated
J. Clin. Endocrinol. Metab. 86, 1410–1417. https://doi.org/10.1210/
recall estimation. Environ. Int. 138, 105623 https://doi.org/10.1016/j.
jcem.86.3.7337.
envint.2020.105623.
Davis, A.P., Wiegers, T.C., Wiegers, J., Grondin, C.J., Johnson, R.J., Sciaky, D.,
IOMC ED (Inter-Organization Programme for the Sound Management of Chemicals
Mattingly, C.J., 2021. CTD anatomy: Analyzing chemical-induced phenotypes and
Environment Directorate). (2017). Guidance document on evaluating and expressing
exposures from an anatomical perspective, with implications for environmental
uncertainty in hazard characterization. Harmonization Project Document 11 – 2nd
health studies. Curr. Res. Toxicol. 2, 128–139. https://doi.org/10.1016/j.
edition (2nd ed.). Geneva, Switzerland: World Health Organization. http://www.
crtox.2021.03.001.
who.int/ipcs/methods/harmonization/areas/hazard_assessment/en/.
Fawcett, J.P., Farquhar, S.J., Thou, T., Shand, B.I., 1997. Oral vanadyl sulphate does not
Keshava, C., Davis, J.A., Stanek, J., Thayer, K.A., Galizia, A., Keshava, N., Gift, J.,
affect blood cells, viscosity or biochemistry in humans. Pharmacol. Toxicol. 80,
Vulimiri, S.V., Woodall, G., Gigot, C., Garcia, K., Greenhalgh, A., Schulz, B.,
202–206.
Volkoff, S., Camargo, K., Persad, A.S., 2020. Application of systematic evidence

31
K.A. Thayer et al. Environment International 169 (2022) 107468

mapping to assess the impact of new research when updating health reference (EPA635R19049). Integrated Risk Information System. Center for Public Health and
values: A case example using acrolein. Environ. Int. 143, 105956 https://doi.org/ Environmental Assessment. Office of Research and Development. U.S.
10.1016/j.envint.2020.105956. Environmental Protection Agency. https://cfpub.epa.gov/ncea/iris_drafts/
McLanahan, E.D., El-Masri, H.A., Sweeney, L.M., Kopylev, L.Y., Clewell, H.J., recordisplay.cfm?deid=345065.
Wambaugh, J.F., Schlosser, P.M., 2012. Physiologically based pharmacokinetic U.S. EPA (U.S. Environmental Protection Agency), 2020a. CompTox Chemicals
model use in risk assessment–Why being published is not enough. Toxicol. Sci. 126, Dashboard. Washington, DC. Retrieved from https://comptox.epa.gov/dashboard.
5–15. https://doi.org/10.1093/toxsci/kfr295. U.S. EPA (U.S. Environmental Protection Agency), 2020b. IRIS assessment plan for oral
NASEM (National Academies of Sciences, Engineering, and Medicine), 2021. Review of exposure to vanadium and compounds (scoping and problem formulation materials).
U.S. EPA’s ORD staff handbook for developing IRIS assessments: 2020 version. (EPA/635/R-20/112). Washington, DC. https://cfpub.epa.gov/ncea/iris_drafts/
Washington, DC: National Academies Press. http://dx.doi.org/10.17226/26289. recordisplay.cfm?deid=348792.
NRC (National Research Council), 1983. Risk assessment in the federal government: U.S. EPA (U.S. Environmental Protection Agency), 2020c. ORD staff handbook for
Managing the process. National Academy Press, Washington, DC http://dx.doi.org/ developing IRIS assessments (public comment draft) [EPA Report]. (EPA/600/R-20/
10.17226/366. 137). Washington, DC: U.S. Environmental Protection Agency, Office of Research
NRC (National Research Council), 2009. Science and decisions: Advancing risk and Development, Center for Public Health and Environmental Assessment. https://
assessment. The National Academies Press, Washington, DC http://dx.doi.org/ cfpub.epa.gov/ncea/iris_drafts/recordisplay.cfm?deid=350086.
10.17226/12209. U.S. EPA, 2020d. PFAS 150 (2020) | Visualizations | Animal Study Evaluation Heatmap.
Radke, E.G., Braun, J.M., Meeker, J.D., Cooper, G.S., 2018. Phthalate exposure and male https://hawcprd.epa.gov/summary/visual/100500560/.
reproductive outcomes: A systematic review of the human epidemiological evidence U.S. EPA (U.S. Environmental Protection Agency), 2020e. PFAS 150 (2020) |
[Review]. Environ. Int. 121, 764–793. https://doi.org/10.1016/j. Visualizations | Available visualizations. https://hawcprd.epa.gov/summary/
envint.2018.07.029. assessment/100500085/visuals/.
Sun, L., Shi, D.J., Gao, X.C., Mi, S.Y., Yu, Y., Han, Q., 2014. The protective effect of U.S. EPA (U.S. Environmental Protection Agency), 2020f. PFAS 150 (2020) |
vanadium against diabetic cataracts in diabetic rat model. Biol. Trace Elem. Res. Visualizations | Main Report Hepatic Weight Oral. https://hawcprd.epa.gov/
158, 219–223. https://doi.org/10.1007/s12011-014-9925-7. summary/data-pivot/assessment/100500085/Main-Report_Hepatic_Weight_Oral/.
Susić, D., Kentera, D., 1986. Effect of chronic vanadate administration on pulmonary U.S. EPA (U.S. Environmental Protection Agency), 2020g. PFAS 150 (2020) |
circulation in the rat. Respiration 49, 68–72. https://doi.org/10.1159/000194861. Visualizations | nonECHA animal study evaluation. https://hawcprd.epa.gov/
U.S. EPA (U.S. Environmental Protection Agency), 2002. A review of the reference dose summary/visual/100500561/.
and reference concentration processes. (EPA/630/P-02/002F). Washington, DC: U.S. U.S. EPA (U.S. Environmental Protection Agency), 2020h. Provisional peer-reviewed
Environmental Protection Agency, Risk Assessment Forum. https://www.epa.gov/ toxicity values for glycidaldehyde [EPA Report]. (EPA/690/R-20/003F).
sites/production/files/2014-12/documents/rfd-final.pdf. Washington, DC: U.S. Environmental Protection Agency, Center for Public Health
U.S. EPA (U.S. Environmental Protection Agency), 2005. Guidelines for carcinogen risk and Environmental Assessment. https://cfpub.epa.gov/ncea/pprtv/recordisplay.
assessment [EPA Report]. (EPA/630/P-03/001B). Washington, DC: U.S. cfm?deid=349744.
Environmental Protection Agency, Risk Assessment Forum. https://www.epa.gov/ U.S. EPA (U.S. Environmental Protection Agency), 2020i. Toxicology testing in the 21st
sites/production/files/2013-09/documents/cancer_guidelines_final_3-25-05.pdf. century (Tox21). Available online at https://ntp.niehs.nih.gov/go/tox21 (accessed
U.S. EPA (U.S. Environmental Protection Agency), 2009. Graphical arrays of chemical- May 4, 2021).
specific health effect reference values for inhalation exposures [EPA Report]. (EPA/ U.S. EPA (U.S. Environmental Protection Agency), 2021. Provisional peer reviewed
600/R-09/061). Research Triangle Park, NC. http://cfpub.epa.gov/ncea/cfm/ toxicity values for trans-Crotonaldehyde [EPA Report]. (EPA/690/R-21/001F).
recordisplay.cfm?deid=211003. Washington, DC: U.S. Environmental Protection Agency, Center for Public Health
U.S. EPA (U.S. Environmental Protection Agency), 2012. Benchmark dose technical and Environmental Assessment. https://cfpub.epa.gov/ncea/pprtv/recordisplay.
guidance. (EPA/100/R-12/001). Washington, DC: U.S. Environmental Protection cfm?deid=351031.
Agency, Risk Assessment Forum. https://www.epa.gov/risk/benchmark-dose- Walker, V.R., Boyles, A.L., Pelch, K.E., Holmgren, S.D., Shapiro, A.J., Blystone, C.R.,
technical-guidance. Devito, M.J., Newbold, R.R., Blain, R., Hartman, P., Thayer, K.A., Rooney, A.A.,
U.S. EPA (U.S. Environmental Protection Agency), 2013. Mn and BTEX reference value 2018. Human and animal evidence of potential transgenerational inheritance of
arrays (final reports). (EPA/600/R-12/047F). Washington, DC. https://cfpub.epa. health effects: An evidence map and state-of-the-science evaluation. Environ. Int.
gov/ncea/risk/recordisplay.cfm?deid=250571. 115, 48–69. https://doi.org/10.1016/j.envint.2017.12.032.
U.S. EPA (U.S. Environmental Protection Agency), 2015. Peer review handbook [EPA Williams, A.J., Lambert, J.C., Thayer, K., Dorne, J., 2021. Sourcing data on chemical
Report] (4th ed.). (EPA/100/B-15/001). Washington, DC: U.S. Environmental properties and hazard data from the US-EPA CompTox Chemicals Dashboard: A
Protection Agency, Science Policy Council. https://www.epa.gov/osa/peer-review- practical guide for human risk assessment [Review]. Environ. Int. 154, 106566
handbook-4th-edition-2015. https://doi.org/10.1016/j.envint.2021.106566.
U.S. EPA (U.S. Environmental Protection Agency), 2018. An umbrella Quality Assurance Woodall, GM; Lipscomb, JC; Schulz, B. (In Press) Visual comparisons of health-based
Project Plan (QAPP) for PBPK models [EPA Report]. (ORD QAPP ID No: B-0030740- toxicity values: I. Overview.
QP-1-1). Research Triangle Park, NC. Yost, E.E., Euling, S.Y., Weaver, J.A., Beverly, B.E.J., Keshava, N., Mudipalli, A.,
U.S. EPA (U.S. Environmental Protection Agency), 2019a. ChemView [Database]. Arzuaga, X., Blessinger, T., Dishaw, L., Hotchkiss, A., Makris, S.L., 2019. Hazards of
Retrieved from https://chemview.epa.gov/chemview. diisobutyl phthalate (DIBP) exposure: A systematic review of animal toxicology
U.S. EPA (U.S. Environmental Protection Agency), 2019b. IRIS assessment plan for studies [Review]. Environ. Int. 125, 579–594. https://doi.org/10.1016/j.
inorganic mercury salts (scoping and problem formulation materials). (EPA/635/R- envint.2018.09.038.
19/149). Washington, DC. https://cfpub.epa.gov/ncea/iris_drafts/recordisplay.cfm? Yost, E.E., Galizia, A., Kapraun, D.F., Persad, A.S., Vulimiri, S.V., Angrish, M., Lee, J.S.,
deid=346843. Druwe, I.L., 2021. Health effects of naphthalene exposure: A systematic evidence
U.S. EPA (U.S. Environmental Protection Agency), 2019c. Systematic review protocol for map and analysis of potential considerations for dose–response evaluation.
the PFBA, PFHxA, PFHxS, PFNA, and PFDA IRIS assessments [EPA Report].

32

You might also like